Patent Application
20190197398
The present disclosure relates to machine learning and, more particularly, to generating a prediction model based on learned latent representations for attributes of entities and content items. SUGGESTED CLASSIFICATION: 709/203; SUGGESTED ART UNIT: 2447.
The Internet allows end-users operating computing devices to request content from many different publishers. Some publishers desire to send additional content items to users who visit their respective websites or who otherwise interact with the publishers. To do so, publishers may rely on a content delivery service that delivers the additional content items over one or more computer networks to computing devices of such users. Some content delivery services have a large database of content items from which to select. It is difficult for a content provider to intelligently select (ahead of time) which of many content items should be delivered in response to each request from a publisher or a computing device of a user.
The approaches described in this section are approaches that could be pursued, but not necessarily approaches that have been previously conceived or pursued. Therefore, unless otherwise indicated, it should not be assumed that any of the approaches described in this section qualify as prior art merely by virtue of their inclusion in this section.
In the drawings:
In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of the present invention. It will be apparent, however, that the present invention may be practiced without these specific details. In other instances, well-known structures and devices are shown in block diagram form in order to avoid unnecessarily obscuring the present invention.
A system and method for using machine learning techniques to predict user selection of content items are provided. Latent representations of attribute values of both content items and entities (users) are automatically learned/generated. For each content item that is identified in a content item selection event, latent representations of values of attributes of the content item are features in a first neural network to generate an output embedding for the content item. Similarly, latent representations of values of attributes of a user that is a target of the content item selection event are features in a second neural network to generate an output embedding for the user. The output embeddings of the content items and the user are used to determine which content item(s) to select for presentation to the user.
This approach to automatically learning latent representations of different attribute values and generating embeddings therefrom improves the accuracy of predicted entity selection rates, resulting in identifying more relevant content items for presentation to requesting entities.
Content providers 112-116 interact with content delivery exchange 120 (e.g., over a network, such as a LAN, WAN, or the Internet) to enable content items to be presented, through publisher 130, to end-users operating client devices 142-146. Thus, content providers 112-116 provide content items to content delivery exchange 120, which in turn selects content items to provide to publisher 130 for presentation to users of client devices 142-146. However, at the time that content provider 112 registers with content delivery exchange 120, neither party may know which end-users or client devices will receive content items from content provider 112.
An example of a content provider includes an advertiser. An advertiser of a product or service may be the same party as the party that makes or provides the product or service. Alternatively, an advertiser may contract with a producer or service provider to market or advertise a product or service provided by the producer/service provider. Another example of a content provider is an online ad network that contracts with multiple advertisers to provide content items (e.g., advertisements) to end users, either through publishers directly or indirectly through content delivery exchange 120.
Although depicted in a single element, content delivery exchange 120 may comprise multiple computing elements and devices, connected in a local network or distributed regionally or globally across many networks, such as the Internet. Thus, content delivery exchange 120 may comprise multiple computing elements, including file servers and database systems.
Publisher 130 provides its own content to client devices 142-146 in response to requests initiated by users of client devices 142-146. The content may be about any topic, such as news, sports, finance, and traveling. Publishers may vary greatly in size and influence, such as Fortune 500 companies, social network providers, and individual bloggers. A content request from a client device may be in the form of a HTTP request that includes a Uniform Resource Locator (URL) and may be issued from a web browser or a software application that is configured to only communicate with publisher 130 (and/or its affiliates). A content request may be a request that is immediately preceded by user input (e.g., selecting a hyperlink on web page) or may initiated as part of a subscription, such as through a Rich Site Summary (RSS) feed. In response to a request for content from a client device, publisher 130 provides the requested content (e.g., a web page) to the client device.
Simultaneously or immediately before or after the requested content is sent to a client device, a content request is sent to content delivery exchange 120. That request is sent (over a network, such as a LAN, WAN, or the Internet) by publisher 130 or by the client device that requested the original content from publisher 130. For example, a web page that the client device renders includes one or more calls (or HTTP requests) to content delivery exchange 120 for one or more content items. In response, content delivery exchange 120 provides (over a network, such as a LAN, WAN, or the Internet) one or more particular content items to the client device directly or through publisher 130. In this way, the one or more particular content items may be presented (e.g., displayed) concurrently with the content requested by the client device from publisher 130.
In response to receiving a content request, content delivery exchange 120 initiates a content item selection event that involves selecting one or more content items (from among multiple content items) to present to the client device that initiated the content request. An example of a content item selection event is an auction.
Content delivery exchange 120 and publisher 130 may be owned and operated by the same entity or party. Alternatively, content delivery exchange 120 and publisher 130 are owned and operated by different entities or parties.
A content item may comprise an image, a video, audio, text, graphics, virtual reality, or any combination thereof. A content item may also include a link (or URL) such that, when a user selects (e.g., with a finger on a touchscreen or with a cursor of a mouse device) the content item, a (e.g., HTTP) request is sent over a network (e.g., the Internet) to a destination indicated by the link. In response, content of a web page corresponding to the link may be displayed on the user's client device.
Examples of client devices 142-146 include desktop computers, laptop computers, tablet computers, wearable devices, video game consoles, and smartphones.
In a related embodiment, system 100 also includes one or more bidders (not depicted). A bidder is a party that is different than a content provider, that interacts with content delivery exchange 120, and that bids for space (on one or more publishers, such as publisher 130) to present content items on behalf of multiple content providers. Thus, a bidder is another source of content items that content delivery exchange 120 may select for presentation through publisher 130. Thus, a bidder acts as a content provider to content delivery exchange 120 or publisher 130. Examples of bidders include AppNexus, DoubleClick, and LinkedIn. Because bidders act on behalf of content providers (e.g., advertisers), bidders create content delivery campaigns and, thus, specify user targeting criteria and, optionally, frequency cap rules, similar to a traditional content provider.
In a related embodiment, system 100 includes one or more bidders but no content providers. However, embodiments described herein are applicable to any of the above-described system arrangements.
Each content provider establishes a content delivery campaign with content delivery exchange 120. A content delivery campaign includes (or is associated with) one or more content items. Thus, the same content item may be presented to users of client devices 142-146. Alternatively, a content delivery campaign may be designed such that the same user is (or different users are) presented different content items from the same campaign. For example, the content items of a content delivery campaign may have a specific order, such that one content item is not presented to a user before another content item is presented to that user.
A content delivery campaign is an organized way to present information to users that qualify for the campaign. Different content providers have different purposes in establishing a content delivery campaign. Example purposes include having users view a particular video or web page, fill out a form with personal information, purchase a product or service, make a donation to a charitable organization, volunteer time at an organization, or become aware of an enterprise or initiative, whether commercial, charitable, or political.
A content delivery campaign has a start date/time and, optionally, a defined end date/time. For example, a content delivery campaign may be to present a set of content items from Jun. 1, 2015 to Aug. 1, 2015, regardless of the number of times the set of content items are presented (“impressions”), the number of user selections of the content items (e.g., click throughs), or the number of conversions that resulted from the content delivery campaign. Thus, in this example, there is a definite (or “hard”) end date. As another example, a content delivery campaign may have a “soft” end date, where the content delivery campaign ends when the corresponding set of content items are displayed a certain number of times, when a certain number of users view the set of content items, select or click on the set of content items, or when a certain number of users purchase a product/service associated with the content delivery campaign or fill out a particular form on a website.
A content delivery campaign may specify one or more targeting criteria that are used to determine whether to present a content item of the content delivery campaign to one or more users. Example factors include date of presentation, time of day of presentation, characteristics of a user to which the content item will be presented, attributes of a computing device that will present the content item, identity of the publisher, etc. Examples of characteristics of a user include demographic information, geographic information (e.g., of an employer), job title, employment status, academic degrees earned, academic institutions attended, former employers, current employer, number of connections in a social network, number and type of skills, number of endorsements, and stated interests. Examples of attributes of a computing device include type of device (e.g., smartphone, tablet, desktop, laptop), geographical location, operating system type and version, size of screen, etc.
For example, targeting criteria of a particular content delivery campaign may indicate that a content item is to be presented to users with at least one undergraduate degree, who are unemployed, who are accessing from South America, and where the request for content items is initiated by a smartphone of the user. If content delivery exchange 120 receives, from a computing device, a request that does not satisfy the targeting criteria, then content delivery exchange 120 ensures that any content items associated with the particular content delivery campaign are not sent to the computing device.
Thus, content delivery exchange 120 is responsible for selecting a content delivery campaign in response to a request from a remote computing device by comparing (1) targeting data associated with the computing device and/or a user of the computing device with (2) targeting criteria of one or more content delivery campaigns. Multiple content delivery campaigns may be identified in response to the request as being relevant to the user of the computing device. Content delivery campaign 120 may select a strict subset of the identified content delivery campaigns from which content items will be identified and presented to the user of the computing device.
Instead of one set of targeting criteria, a single content delivery campaign may be associated with multiple sets of targeting criteria. For example, one set of targeting criteria may be used during one period of time of the content delivery campaign and another set of targeting criteria may be used during another period of time of the campaign. As another example, a content delivery campaign may be associated with multiple content items, one of which may be associated with one set of targeting criteria and another one of which is associated with a different set of targeting criteria. Thus, while one content request from publisher 130 may not satisfy targeting criteria of one content item of a campaign, the same content request may satisfy targeting criteria of another content item of the campaign.
Different content delivery campaigns that content delivery exchange 120 manages may have different charge models. For example, content delivery exchange 120 may charge a content provider of one content delivery campaign for each presentation of a content item from the content delivery campaign (referred to herein as cost per impression or CPM). Content delivery exchange 120 may charge a content provider of another content delivery campaign for each time a user interacts with a content item from the content delivery campaign, such as selecting or clicking on the content item (referred to herein as cost per click or CPC). Content delivery exchange 120 may charge a content provider of another content delivery campaign for each time a user performs a particular action, such as purchasing a product or service, downloading a software application, or filling out a form (referred to herein as cost per action or CPA). Content delivery exchange 120 may manage only campaigns that are of the same type of charging model or may manage campaigns that are of any combination of the three types of charging models.
A content delivery campaign may be associated with a resource budget that indicates how much the corresponding content provider is willing to be charged by content delivery exchange 120, such as $100 or $5,200. A content delivery campaign may also be associated with a bid amount that indicates how much the corresponding content provider is willing to be charged for each impression, click, or other action. For example, a CPM campaign may bid five cents for an impression, a CPC campaign may bid five dollars for a click, and a CPA campaign may bid five hundred dollars for a conversion (e.g., a purchase of a product or service).
As mentioned previously, a content item selection event is when multiple content items (e.g., from different content delivery campaigns) are considered and a subset selected for presentation on a computing device in response to a request. Thus, each content request that content delivery exchange 120 receives triggers a content item selection event.
For example, in response to receiving a content request, content delivery exchange 120 analyzes multiple content delivery campaigns to determine whether attributes associated with the content request (e.g., attributes of a user that initiated the content request, attributes of a computing device operated by the user, current date/time) satisfy targeting criteria associated with each of the analyzed content delivery campaigns. If so, the content delivery campaign is considered a candidate content delivery campaign. One or more filtering criteria may be applied to a set of candidate content delivery campaigns to reduce the total number of candidates.
As another example, users are assigned to content delivery campaigns (or specific content items within campaigns) “off-line”; that is, before content delivery exchange 120 receives a content request that is initiated by the user. For example, when a content delivery campaign is created based on input from a content provider, one or more computing components may compare the targeting criteria of the content delivery campaign with attributes of many users to determine which users are to be targeted by the content delivery campaign. If a user's attributes satisfy the targeting criteria of the content delivery campaign, then the user is assigned to a target audience of the content delivery campaign. Thus, an association between the user and the content delivery campaign is made. Later, when a content request that is initiated by the user is received, all the content delivery campaigns that are associated with the user may be quickly identified, in order to avoid real-time (or on-the-fly) processing of the targeting criteria. Some of the identified campaigns may be further filtered based on, for example, the campaign being deactivated or terminated, the device that the user is operating being of a different type (e.g., desktop) than the type of device targeted by the campaign (e.g., mobile device).
A final set of candidate content delivery campaigns is ranked based on one or more criteria, such as predicted click-through rate (which may be relevant only for CPC campaigns), effective cost per impression (which may be relevant to CPC, CPM, and CPA campaigns), and/or bid price. Each content delivery campaign may be associated with a bid price that represents how much the corresponding content provider is willing to pay (e.g., content delivery exchange 120) for having a content item of the campaign presented to an end-user or selected by an end-user. Different content delivery campaigns may have different bid prices. Generally, content delivery campaigns associated with relatively higher bid prices will be selected for displaying their respective content items relative to content items of content delivery campaigns associated with relatively lower bid prices. Other factors may limit the effect of bid prices, such as objective measures of quality of the content items (e.g., actual click-through rate (CTR) and/or predicted CTR of each content item), budget pacing (which controls how fast a campaign's budget is used and, thus, may limit a content item from being displayed at certain times), frequency capping (which limits how often a content item is presented to the same person), and a domain of a URL that a content item might include.
An example of a content item selection event is an advertisement auction, or simply an “ad auction.”
In one embodiment, content delivery exchange 120 conducts one or more content item selection events. Thus, content delivery exchange 120 has access to all data associated with making a decision of which content item(s) to select, including bid price of each campaign in the final set of content delivery campaigns, an identity of an end-user to which the selected content item(s) will be presented, an indication of whether a content item from each campaign was presented to the end-user, a predicted CTR of each campaign, a CPC or CPM of each campaign.
In another embodiment, an exchange that is owned and operated by an entity that is different than the entity that owns and operates content delivery exchange 120 conducts one or more content item selection events. In this latter embodiment, content delivery exchange 120 sends one or more content items to the other exchange, which selects one or more content items from among multiple content items that the other exchange receives from multiple sources. In this embodiment, content delivery exchange 120 does not know (a) which content item was selected if the selected content item was from a different source than content delivery exchange 120 or (b) the bid prices of each content item that was part of the content item selection event. Thus, the other exchange may provide, to content delivery exchange 120 (or to a performance simulator described in more detail herein), information regarding one or more bid prices and, optionally, other information associated with the content item(s) that was/were selected during a content item selection event, information such as the minimum winning bid or the highest bid of the content item that was not selected during the content item selection event.
Content delivery exchange 120 tracks one or more types of user interactions across client devices 142-146 (and other client devices not depicted). For example, content delivery exchange 120 determines whether a content item that content delivery exchange 120 delivers is presented at (e.g., displayed by or played back at) a client device. Such a “user interaction” is referred to as an “impression.” As another example, content delivery exchange 120 determines whether a content item that exchange 120 delivers is selected by a user of a client device. Such a “user interaction” is referred to as a “click.” Content delivery exchange 120 stores such data as user interaction data, such as an impression data set and/or a click data set.
For example, content delivery exchange 120 receives impression data items, each of which is associated with a different instance of an impression and a particular content delivery campaign. An impression data item may indicate a particular content delivery campaign (e.g., a campaign identifier), a content provider of the campaign (e.g., a content provider identifier), a specific content item (e.g., a content item identifier), a date of the impression, a time of the impression, a particular publisher or source (e.g., onsite v. offsite), a particular client device that displayed the specific content item, and/or a user identifier of a user that operates the particular client device. Thus, if content delivery exchange 120 manages multiple content delivery campaigns, then different impression data items may be associated with different content delivery campaigns. One or more of these individual data items may be encrypted to protect privacy of the end-user. An impression data item may contain a content item identifier that is used (later) by content delivery exchange 120 to look up a campaign identifier (that uniquely identifies a content delivery campaign to which the content item belongs) and/or a content provider identifier (that uniquely identifies a content provider that provided or created the campaign).
Similarly, a click data item may indicate a particular content delivery campaign, a specific content item, a date of the user selection, a time of the user selection, a particular publisher or source (e.g., onsite v. offsite), a particular client device that displayed the specific content item, and/or a user identifier of a user that operates the particular client device. If impression data items are generated and processed properly, a click data item should be associated with an impression data item that corresponds to the click data item.
At block 210, a request for one or more content items is received. The request is initiated by a computing device (e.g., client device 110) that is operated by a requesting entity or user. The request may have been generated and transmitted when the computing device loaded a web page that includes code for generating the request. The web page may be provided by a server that is in the same domain or network as content delivery exchange 120.
At block 220, a content item selection event is initiated and multiple content items are identified. Content items are associated with targeting criteria and, in order to be identified in block 220, the targeting criteria of a content item should be satisfied (at least partially). If a content delivery campaign includes multiple content items, then the multiple content items may share the same targeting criteria. Alternatively, two or more content items belonging to the same content delivery campaign may be associated with different targeting criteria relative to each other. If no targeting criteria of any content item is satisfied, then default content items may be identified or randomly selected.
At block 230, multiple machine-learned embeddings of the user are identified and, for each identified content item, multiple machine-learned embeddings of the content item are identified. A machine-learned embedding is a vector of real numbers and represents a word or identifier. How embeddings are generated is described in more detail below.
Each machine-learning embedding corresponds to a value of an attribute (or attribute value). Example attributes of a content item include content provider identifier (that uniquely identifies a content provider that provided the content delivery campaign that includes the content item), campaign identifier (that uniquely identifies the content delivery campaign), and content item identifier (that uniquely identifies the content item). Each identifier may be globally unique or at least unique within the attribute to which the identifier pertains. Additionally or alternatively, an identifier may be a name (e.g., company=LinkedIn) or may be an identifier (e.g., whether numeric or alphanumeric) that has been mapped to the name (e.g., company=12345).
Example attributes of a user include an employer, a job title, a skill, and industry. Again, the corresponding attribute values may be actual names (e.g., “Software Engineer” for job title or “Finance” for industry) or may be identifiers to which the names have been mapped.
Block 230 may involve first identifying attribute values of a content item (e.g., from a content item database) and attribute values of a user (e.g., from a profile database) and then using one or more mappings or tables to identify, for each identified attribute value, a machine-learned embedding that correspond to that attribute value.
Input embeddings 312-318 are learned embeddings for different attribute values of a user. Input embedding 312 corresponds to a particular employer or organization, input embedding 314 corresponds to a particular job title, input embedding 316 corresponds to a particular skill, and input embedding 318 corresponds to a particular industry. Other embodiments may include more or less embeddings. For example, one embodiment may exclude industry as an attribute while another embodiment may include academic institution and degree earned as additional attributes in which embeddings will be learned. In an embodiment, a vector size of an input embedding is between five and fifteen dimensions or values.
Although input embeddings 302-318 are depicted as being vectors of size five, actual embeddings may be vectors of a larger size, such as ten.
At block 240, for each identified content item, the multiple embeddings of the content item are combined to create an “initial” content item-level embedding. The combining may involve concatenating the individual embeddings of the attributes of the content item.
In
At block 250, for each identified content item, the corresponding initial content item-level embedding is input to a first neural network that comprises an input layer, one or more hidden layers, and an output layer. The first neural network may be a fully-connected network. In an embodiment, the first neural network has two hidden layers. The output layer produces “final” content item-level embedding, which is a vector of real numbers of a particular size. In an embodiment, the vector size of the final embedding is between 150 and 350.
In
At block 260, the multiple embeddings associated with the user are combined to create an “initial” user-level embedding. The combining may involve concatenating the individual embeddings of the attributes of the user. The size of an initial user-level embedding may be larger or smaller than each content item-level embedding. Such a difference in size may be due to the number of content item attributes that are considered (e.g., 3) being different than the number of user attributes that are considered (e.g., 4). Alternatively, the size of individual embeddings of content item attributes (e.g., 8) may be different than the size of individual embeddings of user attributes (e.g., 10).
In
At block 270, the initial user-level embedding is input into a second neural network that also comprises an input layer, one or more hidden layers, and an output layer. The second neural network may also be a fully-connected network. The output layer produces a “final” user-level embedding, which is a vector of the same size as the vector produced by the output layer of the first neural network. While the second neural network is utilized once for each content item selection event, the first neural network is utilized multiple times for each content item selection event, once for each identified (or candidate) content item.
In
At block 280, for each identified content item, an operation on the output of the first neural network (i.e., final content item-level embedding) and the output of the second neural network (i.e., the final user-level embedding) is performed to generate a result. The operation may be a dot product, difference, or summation. The more similar the outputs of the respective neural networks, the more likely the corresponding user will select (or otherwise interact with) the corresponding content item. Any similarity can be used as a signal for down-stream interaction (e.g., selection). As a specific example, 1/(1−ê−(L1*L2)) is computed, where L1 is the final content item-level embedding produced by the first neural network, L2 is the final user-level embedding produced by the second neural network, and ‘*’ is a dot product operation. The result of this computation reflects a probability that the user corresponding to L2 will select the content item corresponding to L1.
In
At block 290, based on the generated results, one or more of the identified content items are selected for delivery to the computing device of the user. In some cases, a content item selection event may result in selecting a single content item for presentation, while, in other cases, a content item selection event may result in selecting multiple content items for presentation. The results of block 280 may be one of many factors that are considered when selecting a content item. For example, a bid price of each identified content item may be another factor in determining which content item(s) to select for presentation.
In an embodiment, final content item-level embeddings and/or final user-level embeddings are stored and retrieved later when the corresponding content items and/or users are identified in future content item selection events. For example, if content item A is identified in a first content item selection event and a final content item-level embedding is generated for content item A, then that final content item-level embedding is stored in association within content item A. Later, during a second content item selection event, content item A is identified again and the final content item-level embedding is retrieved from storage without having to construct an initial content item-level embedding and feed that embedding into the first neural network to generate the final content item-level embedding. Thus, blocks 230-270 may be replaced with retrieval, from storage, of a final user-level embedding and of final content item-level embeddings of the content items identified in block 220.
In an embodiment, final content item-level embeddings and/or final user-level embeddings are generated prior to (rather than in response to) a content request, processing of which would require one or more of the final embeddings. For example, a final user-level embedding is generated for each of multiple users soon after a machine learned prediction model (comprising multiple artificial neural networks) is generated. The multiple users may be users who are known to have selected a content item in the recent past or otherwise initiated content item selection events in the recent past. As another example, a final content item-level embedding is generated for each of multiple content items soon after the machine-learned prediction model is generated. The multiple content items may be content items that have been candidates in content item selection events in the recent past.
An embedding is a vector of real numbers. “Embedding” is a name for a set of feature learning techniques where words or identifiers are mapped to vectors of real numbers. Conceptually, embedding involves a mathematical embedding from a space with one dimension per word/phrase (or identifier) to a continuous vector space.
One method to generate embeddings includes artificial neural networks. In the context of linguistics, word embedding, when used as the underlying input representation, have been shown to boost performance in natural language processing (NLP) tasks, such as syntactic parsing and sentiment analysis. Word embedding aims to quantify and categorize semantic similarities between linguistic items based on their distributional properties in large samples of language data. The underlying idea that a word is characterized by “the company it keeps.”
In an embodiment, in the context of content item selection, an embedding is learned for each of multiple content item attribute values and each of multiple user attribute values. Such attribute values may be string values or numeric identifiers. For example, a content item attribute includes content provider, which, for a particular content item, may be a name (e.g., string of non-numeric characters) of the content provider (e.g., “Company X”) or an identifier (e.g., “435256”) that uniquely identifies the content provider.
Each embedding represents something different. For example, an embedding for a particular employer (which embedding is used to generate an initial user-level embedding) represents behavior of employees of the particular employer when responding to selectable content items (e.g., whether clicking the content items or not). Similarly, an embedding for a particular job title (which embedding is used to generate an initial user-level embedding) represents behavior of users with that particular job title when responding to selectable content items. As another example, an embedding for a particular content provider (which embedding is used to generate an initial content item-level embedding) represents user behavior towards selectable content items provided by the particular content provider. Similarly, an embedding for a particular content delivery campaign (which embedding is used to generate an initial content item-level embedding) represents user behavior towards selectable content items that belong to that particular content delivery campaign.
The training data that is used to generate or “learn” embeddings for different attribute values comprises a portion of the user interaction data described previously. In order to generate the training data, the original user interaction data may have been augmented with additional information and/or may have been filtered to remove unnecessary data, such as timestamp data. For example, given a click data item that includes a member identifier, the member identifier is used to look up, in a profile database, a profile and retrieve one or more data items from the profile, such as one or more most recent job titles, one or more skills, and an industry. If the retrieved attribute values are names and not identifiers, then each retrieved attribute value name may be used to lookup, in a mapping (e.g., “‘Software Engineer’→87654”), a unique internal identifier that is mapped to the retrieved attribute value name. As another example, given an impression data item that includes a content item identifier, the content item identifier is used to look up, in a content item database, a record that includes a campaign name (or identifier) and/or a content provider name (or identifier).
Thus, each training instance indicates multiple content item-related attribute values and multiple user-related attribute values. Content item-related attribute values include a content item identifier that uniquely identifies a content item, a content delivery campaign identifier that uniquely identifies a content delivery campaign to which the content item belongs, a content provider identifier that uniquely identifies a content provider that initiated or created the content delivery campaign.
User-related attribute values include: a user identifier that uniquely identifies a user (e.g., a member of a social network), one or more employer identifiers, each of which uniquely identifies an employer that the user may have specified in his/her profile; one or more job title identifiers, each of which uniquely identifies a job title that the user may have specified in his/her profile; one or more skill identifiers, each of which uniquely identifies a skill that the user may have specified in his/her profile; and an industry identifier that uniquely identifies an industry that the user may have specified in his/her profile or that may have been derived based on a job title (and/or other information) associated with the user.
Each training instance also indicates whether the indicated content item was selected or otherwise interacted with by the indicated user. For example, a ‘1’ may indicate that the corresponding user “clicked” on the corresponding content item and a ‘0’ may indicate that the corresponding user did not click on the corresponding content item. In practice, very few content items are selected by a user, such as under 0.4%. One way to deal with imbalanced labels is to downsample the negative samples. A more effective way to deal with imbalanced labels is to upsample the positive samples. Additionally or alternatively, positive samples may be weighted more than negative samples through weighted regularization, weighted costs functions or other approaches.
The user interaction data upon which the training data is based may be limited to user interaction data that was generated during a certain time period, such as the last fourteen days.
In training multiple artificial neural networks, embeddings for attribute values that are indicated in the training data may be initialized to random numbers at the beginning. During the training process, each embedding is continuously modified until the embedding “stabilizes”, such that the object value that is being optimized stops significantly improving. Training may be performed in small batches and embeddings may be updated after each batch. A stabilized embedding becomes a “final” embedding for the corresponding attribute value. A final embedding and its corresponding attribute value may be stored in a mapping or table of multiple final embeddings. For example, one table may store associations between final embeddings and attribute values pertaining to content providers and another table may store associations between final embeddings and attribute values pertaining to job titles.
The training process involves gradient descent and backpropagation. Gradient descent is an iterative optimization algorithm for finding the minimum of a function; in this case, a loss function. Backpropagation is a method used in artificial neural networks to calculate the error contribution of each neuron after a batch of data is processed. In the context of learning, backpropagation is used by a gradient descent optimization algorithm to adjust the weight of neurons (or nodes) in a neural network by calculating the gradient of the loss function. Backpropagation is also referred to as the “backward propagation of errors” because the error is calculated at the output and distributed back through the network layers. For models involving embeddings, there is an implicit input layer that is often not mentioned. The embeddings are actually a layer by themselves and backpropagation goes all the way back to the embedding layer. The input layer maps inputs to the embedding layer. Backpropagation begins at the final (output) layer that generates the probabilities and is applied per batch. Batch size depends on several factors, including the available memory on the computing device or GPU.
For example, employer “LinkedIn” may be mapped to “employer=12345”. For each training instance (e.g., impression or click) in which the identified member lists “LinkedIn” as their employer, the random vector for “employer=12345” would be the same. The initial random vector for employer=12345 is modified after the first training instance (or after a batch of training instances), the modified vector is retained, and the modified vector is used the next time employer=12345 appears in a training instance.
After generating embeddings for different attribute values of different attributes during the training process, the embeddings are associated with their respective attribute values. For example, an embedding for a first content provider is stored in association with the first content provider (such as a unique content provider identifier). Similarly, an embedding for a particular skill (e.g., “Cloud Computing”, which may be mapped for a particular internal identifier that represents that skill) is stored in association with that particular skill.
Later, when a content request is received that is initiated by a particular user, embeddings of attribute values of the particular user are retrieved, along with embeddings of attribute values of one or more content items that are candidates for presentation to the particular user. For example, a content request may include a user/member identifier that is used to lookup a profile of the particular user in a profile database. As part of the lookup, certain attribute names are used in the lookup, such as “Job Title”, “Employer”, etc. The corresponding attribute values are retrieved from the profile. One or more mappings of attribute values to their respective embeddings are accessed to determine the embeddings of the retrieved attribute values. As noted previously, there may be a separate mapping or table for each attribute. For example, one mapping is used for employer while another mapping is used for job title. The retrieved embeddings are then combined (e.g., concatenated) to generate an initial user-level embedding, which is input to the appropriate artificial neural network for users in order to generate, as output, a final user-level embedding.
On the content item side, a content request initiates a content item selection event where multiple content items from different content delivery campaigns are identified as candidate content items for presentation to a user. For each candidate content item, attribute values of the candidate content item are identified and, for each attribute value, an embedding is retrieved. Then, an initial content item-level embedding is generated for a candidate content item based on (e.g., by concatenating) the individual embeddings retrieved for the candidate content item. The content item-level embedding is then input into the appropriate artificial neural network for content items in order to generate, as output, a final content item-level embedding.
For each final content item-level embedding, that final content item-level embedding and the final user-level embedding are input to a function that performs one or more operations and generates a result. Thus, a different result is generated for each content item. The results are used to select a subset of the candidate content items. For example, the greater the value of a result, the greater the probability of the corresponding user selecting the corresponding content item. The value of each result may be one of multiple features that are considered in selecting a subset of the candidate content items. For example, the generated results may be input into another machine-learned prediction model that is used to select a subset of the candidate content items.
In an embodiment, a user is associated with multiple values of a particular attribute. For example, a user might have been employed by multiple companies (whether concurrently or serially over time), might have multiple job titles (whether concurrently or serially over time), and might have multiple skills. When generating a user-level embedding, if a user has multiple values of a particular attribute, then the embeddings associated with the multiple values are combined before combining (e.g., concatenating) the embedding associated with that particular attribute with embeddings associated with attribute values of other attributes. For example, a user has been employed by multiple employers over time. An embedding associated with each employer is identified. Each embedding is a vector comprising multiple ordered entries, each entry containing a real number.
The maximum, average, median, or minimum of each entry relative to other entries in other embeddings of the same index is determined. For example, for a result embedding that is generated based on a set of embeddings of a particular attribute, the first entry in the result embedding will contain the maximum value of the first entries of the embeddings in the set of embeddings; the second entry in the result embedding will contain the maximum value of the second entries of the embeddings in the set of embeddings; the third entry in the result embedding will contain the maximum value of the third entries of the embeddings in the set of embeddings; and so forth. Such a process is referred to as “max pooling.” A similar process may be performed where, instead of finding the maximum value, the median value or the mean value is computed for each entry in the result embedding.
In an embodiment, embeddings for one or more attribute values are missing for a content item or a user. An embedding may not be available for an attribute value if an embedding has not yet been learned for the attribute value. For example, no embedding has yet been learned for a new content item that was created in the last 24 hours. Similarly, no embedding has yet been learned for a new content delivery campaign. As another example, a particular skill, job title, or employer may be new, in which case no embedding will have been learned for that attribute value.
In this embodiment, if no embedding exists for an attribute value, then a random embedding is generated. Alternatively, if the missing embedding is for a new content item, then embeddings of other content items from the same content delivery campaign (to which the new content item belongs) may be combined (e.g., averaged, median determined, or maximum determined) to generate a combined content item embedding. The combined content item embedding is used for the new content item until a machine-learned embedding is generated for the new content item. The combined content item embedding is then combined (e.g., concatenated) with one or more other attribute values of the content item to generate a content item-level embedding.
If a missing embedding is for a new content delivery campaign, then embeddings of other content delivery campaigns from the same content provider (that has initiated the new content delivery campaign) may be combined (e.g., averaged, median determined, or maximum determined) to generate a combined campaign embedding. The combined campaign embedding is used until a machine-learned embedding is generated for the new content item. The combined campaign embedding is then combined (e.g., concatenated) with one or more other attribute values of the corresponding content item to generate a content item-level embedding.
If a missing embedding is for a new content provider, then embeddings of other content providers may be combined (e.g., averaged, median determined, or maximum determined) to generate a combined content provider embedding. The combined content provider embedding is used until a machine-learned embedding is generated for the new content provider.
If a missing embedding is for a new job title, then embeddings of job titles that are considered similar to the new job title may be combined. A similar job title may be one that has similar words or meanings as the new job title. A similar process for new skills may be followed.
In some scenarios, a user might not fill in his/her profile with sufficient information, such that one or more attribute values are missing. For example, a user might not specify any skills (or might specify very few skills) in her profile. Skills of the user may be inferred by determining the most frequently specified skills in profiles of users (a) with the same job title, (b) at the same employer, and/or (c) who are connected to the user in a social network. The top N (e.g., two or three) of those skills are associated with the user (though not included in the user's profile) and embeddings of those top N skills are retrieved and user to generate a user-level embedding for the skill attribute.
As another example, a user might not specify an industry in his/her profile. An industry of the user may be inferred by determining the most common specified industry in profiles of users (a) with the same job title, (b) at the same employer, and/or (c) who are connected to the user in a social network. The top N (e.g., one, two, or three) of those industries are associated with the user (though not included in the user's profile) and embeddings of those top N industries are retrieved and user to generate a user-level embedding for the industry attribute.
According to one embodiment, the techniques described herein are implemented by one or more special-purpose computing devices. The special-purpose computing devices may be hard-wired to perform the techniques, or may include digital electronic devices such as one or more application-specific integrated circuits (ASICs) or field programmable gate arrays (FPGAs) that are persistently programmed to perform the techniques, or may include one or more general purpose hardware processors programmed to perform the techniques pursuant to program instructions in firmware, memory, other storage, or a combination. Such special-purpose computing devices may also combine custom hard-wired logic, ASICs, or FPGAs with custom programming to accomplish the techniques. The special-purpose computing devices may be desktop computer systems, portable computer systems, handheld devices, networking devices or any other device that incorporates hard-wired and/or program logic to implement the techniques.
For example,
Computer system 400 also includes a main memory 406, such as a random access memory (RAM) or other dynamic storage device, coupled to bus 402 for storing information and instructions to be executed by processor 404. Main memory 406 also may be used for storing temporary variables or other intermediate information during execution of instructions to be executed by processor 404. Such instructions, when stored in non-transitory storage media accessible to processor 404, render computer system 400 into a special-purpose machine that is customized to perform the operations specified in the instructions.
Computer system 400 further includes a read only memory (ROM) 408 or other static storage device coupled to bus 402 for storing static information and instructions for processor 404. A storage device 410, such as a magnetic disk, optical disk, or solid-state drive is provided and coupled to bus 402 for storing information and instructions.
Computer system 400 may be coupled via bus 402 to a display 412, such as a cathode ray tube (CRT), for displaying information to a computer user. An input device 414, including alphanumeric and other keys, is coupled to bus 402 for communicating information and command selections to processor 404. Another type of user input device is cursor control 416, such as a mouse, a trackball, or cursor direction keys for communicating direction information and command selections to processor 404 and for controlling cursor movement on display 412. This input device typically has two degrees of freedom in two axes, a first axis (e.g., x) and a second axis (e.g., y), that allows the device to specify positions in a plane.
Computer system 400 may implement the techniques described herein using customized hard-wired logic, one or more ASICs or FPGAs, firmware and/or program logic which in combination with the computer system causes or programs computer system 400 to be a special-purpose machine. According to one embodiment, the techniques herein are performed by computer system 400 in response to processor 404 executing one or more sequences of one or more instructions contained in main memory 406. Such instructions may be read into main memory 406 from another storage medium, such as storage device 410. Execution of the sequences of instructions contained in main memory 406 causes processor 404 to perform the process steps described herein. In alternative embodiments, hard-wired circuitry may be used in place of or in combination with software instructions.
The term “storage media” as used herein refers to any non-transitory media that store data and/or instructions that cause a machine to operate in a specific fashion. Such storage media may comprise non-volatile media and/or volatile media. Non-volatile media includes, for example, optical disks, magnetic disks, or solid-state drives, such as storage device 410. Volatile media includes dynamic memory, such as main memory 406. Common forms of storage media include, for example, a floppy disk, a flexible disk, hard disk, solid-state drive, magnetic tape, or any other magnetic data storage medium, a CD-ROM, any other optical data storage medium, any physical medium with patterns of holes, a RAM, a PROM, and EPROM, a FLASH-EPROM, NVRAM, any other memory chip or cartridge.
Storage media is distinct from but may be used in conjunction with transmission media. Transmission media participates in transferring information between storage media. For example, transmission media includes coaxial cables, copper wire and fiber optics, including the wires that comprise bus 402. Transmission media can also take the form of acoustic or light waves, such as those generated during radio-wave and infra-red data communications.
Various forms of media may be involved in carrying one or more sequences of one or more instructions to processor 404 for execution. For example, the instructions may initially be carried on a magnetic disk or solid-state drive of a remote computer. The remote computer can load the instructions into its dynamic memory and send the instructions over a telephone line using a modem. A modem local to computer system 400 can receive the data on the telephone line and use an infra-red transmitter to convert the data to an infra-red signal. An infra-red detector can receive the data carried in the infra-red signal and appropriate circuitry can place the data on bus 402. Bus 402 carries the data to main memory 406, from which processor 404 retrieves and executes the instructions. The instructions received by main memory 406 may optionally be stored on storage device 410 either before or after execution by processor 404.
Computer system 400 also includes a communication interface 418 coupled to bus 402. Communication interface 418 provides a two-way data communication coupling to a network link 420 that is connected to a local network 422. For example, communication interface 418 may be an integrated services digital network (ISDN) card, cable modem, satellite modem, or a modem to provide a data communication connection to a corresponding type of telephone line. As another example, communication interface 418 may be a local area network (LAN) card to provide a data communication connection to a compatible LAN. Wireless links may also be implemented. In any such implementation, communication interface 418 sends and receives electrical, electromagnetic or optical signals that carry digital data streams representing various types of information.
Network link 420 typically provides data communication through one or more networks to other data devices. For example, network link 420 may provide a connection through local network 422 to a host computer 424 or to data equipment operated by an Internet Service Provider (ISP) 426. ISP 426 in turn provides data communication services through the world wide packet data communication network now commonly referred to as the “Internet” 428. Local network 422 and Internet 428 both use electrical, electromagnetic or optical signals that carry digital data streams. The signals through the various networks and the signals on network link 420 and through communication interface 418, which carry the digital data to and from computer system 400, are example forms of transmission media.
Computer system 400 can send messages and receive data, including program code, through the network(s), network link 420 and communication interface 418. In the Internet example, a server 430 might transmit a requested code for an application program through Internet 428, ISP 426, local network 422 and communication interface 418.
The received code may be executed by processor 404 as it is received, and/or stored in storage device 410, or other non-volatile storage for later execution.
In the foregoing specification, embodiments of the invention have been described with reference to numerous specific details that may vary from implementation to implementation. The specification and drawings are, accordingly, to be regarded in an illustrative rather than a restrictive sense. The sole and exclusive indicator of the scope of the invention, and what is intended by the applicants to be the scope of the invention, is the literal and equivalent scope of the set of claims that issue from this application, in the specific form in which such claims issue, including any subsequent correction.
