Patent Application
20200226188
This disclosure relates to the field of computer networks, and more particularly to recommending network-based content to users.
Computer networks, such as the Internet, provide users access to a variety of content, such as text-based documents (e.g., web pages, news articles, etc.), video, audio, or other types of content. Network-based content is generally referred to herein as electronic documents or documents. A typical user accesses the documents through an application (e.g., an email client, a web browser, etc.) on his/her computing device. For example, a web browser accesses web pages over a network connection, and displays the web pages to the user for viewing/reading. Each day, the volume of new documents available to a user over a computer network is massive, so systems may be developed that recommend certain documents to the user. For example, a web browser may generate a profile for a user based on a browsing history of the user, and recommend web pages to the user based on the profile. However, a static profile such as this may quickly become outdated as new topics, content, and reading patterns evolve. Therefore, it is desirable to identify improved ways of recommending documents to users.
Embodiments described herein provide a system, method, and software that recommend documents of interest to users. For example, a system as described herein uses a multi-step approach to construct a refined reading matrix that is used to provide recommendations of new documents that become available to users. For one step, the system constructs an estimated reading matrix that specifies a reading score for past documents (e.g., more than a month old) and for new or recent documents (e.g., published within the last month). The reading scores for the past documents are true scores, which means that the scores are based on actual reading by users. The reading scores for the recent documents are estimated by the system, such as using a cosine-averaged algorithm. For the second step, the system uses matrix decomposition on the estimated reading matrix to generate the refined reading matrix. This approach is advantageous in terms of processing speed so that real-time recommendations may be made to users. Another advantage is that the approach is adaptive in that the refined reading matrix evolves based on how the users read and/or score the recent documents, which subsequently transition to past documents over time. Thus, the refined reading matrix may be used to deliver more relevant documents to each user.
One embodiment comprises a recommendation system that comprises first circuitry configured to identify recent documents published within a time period preceding a present date, and to identify historical information indicating consumption by users of past documents that were published prior to the time period. The recommendation system further comprises second circuitry configured to generate a historical reading matrix based on the historical information, where the historical reading matrix has rows for the users, columns for the past documents, and entries that represent actual reading scores for the past documents. The second circuitry is further configured to generate a vector matrix for the recent documents and the past documents. The recommendation system further comprises third circuitry configured to generate an estimated reading matrix based on the historical reading matrix and the vector matrix, where the estimated reading matrix has rows for the users, columns for the past documents and the recent documents, and first entries that represent the actual reading scores for the past documents. In generating the estimated reading matrix, the third circuitry is configured to calculate estimated reading scores for the recent documents based on the vector matrix, and to populate second entries of the estimated reading matrix corresponding to the recent documents with the estimated reading scores. The recommendation system further comprises fourth circuitry configured to perform a factorization on the estimated reading matrix to generate a refined reading matrix. The recommendation system further comprises fifth circuitry configured to generate recommendations of the recent documents to the users based on the refined reading matrix.
In another embodiment, the third circuitry is configured to calculate the estimated reading scores for the recent documents using a cosine-averaged algorithm.
In another embodiment, the fourth circuitry is configured to use Nonnegative Matrix Factorization (NMF) to generate the refined reading matrix from the estimated reading matrix.
In another embodiment, the fourth circuitry is configured to use Singular Value Decomposition (SVD) to generate the refined reading matrix from the estimated reading matrix.
In another embodiment, the fourth circuitry is configured to use Lower-Upper (LU) decomposition to generate the refined reading matrix from the estimated reading matrix.
In another embodiment, the fifth circuitry is configured to present the recommendations of the recent documents to at least one of the users through a Graphical User Interface (GUI).
Another embodiment comprises a method of recommending documents to users. The method comprises identifying recent documents published within a time period preceding a present date, and identifying historical information indicating consumption by users of past documents that were published prior to the time period. The method further comprises generating a historical reading matrix based on the historical information, where the historical reading matrix has rows for the users, columns for the past documents, and entries that represent actual reading scores for the past documents. The method further comprises generating a vector matrix for the recent documents and the past documents. The method further comprises generating an estimated reading matrix based on the historical reading matrix and the vector matrix, where the estimated reading matrix has rows for the users, columns for the past documents and the recent documents, and first entries that represent the actual reading scores for the past documents. In generating the vector matrix, the method further comprises calculating estimated reading scores for the recent documents based on the vector matrix, and populating second entries of the estimated reading matrix corresponding to the recent documents with the estimated reading scores. The method further comprises performing a factorization on the estimated reading matrix to generate a refined reading matrix, and generating recommendations of the recent documents to the users based on the refined reading matrix.
In another embodiment, calculating the estimated reading scores for the recent documents based on the vector matrix comprises calculating the estimated reading scores for the recent documents using a cosine-averaged algorithm.
In another embodiment, performing the factorization on the estimated reading matrix to generate the refined reading matrix comprises performing Nonnegative Matrix Factorization (NMF) to generate the refined reading matrix from the estimated reading matrix.
In another embodiment, performing the factorization on the estimated reading matrix to generate the refined reading matrix comprises performing Singular Value Decomposition (SVD) to generate the refined reading matrix from the estimated reading matrix.
In another embodiment, performing the factorization on the estimated reading matrix to generate the refined reading matrix comprises performing Lower-Upper (LU) decomposition to generate the refined reading matrix from the estimated reading matrix.
Another embodiment comprises a recommendation system that includes a means for identifying recent documents published within a time period preceding a present date, and for identifying historical information indicating consumption by users of past documents that were published prior to the time period. The recommendation system further comprises a means for generating a historical reading matrix based on the historical information, where the historical reading matrix has rows for the users, columns for the past documents, and entries that represent actual reading scores for the past documents. The recommendation system further comprises a means for generating a vector matrix for the recent documents and the past documents. The recommendation system further comprises a means for generating an estimated reading matrix based on the historical reading matrix and the vector matrix, where the estimated reading matrix has rows for the users, columns for the past documents and the recent documents, and first entries that represent the actual reading scores for the past documents. The recommendation system further comprises a means for calculating estimated reading scores for the recent documents based on the vector matrix, and for populating second entries of the estimated reading matrix corresponding to the recent documents with the estimated reading scores. The recommendation system further comprises a means for performing a factorization on the estimated reading matrix to generate a refined reading matrix, and for generating recommendations of the recent documents to the users based on the refined reading matrix.
Other embodiments may include computer readable media, other systems, or other methods as described below.
The above summary provides a basic understanding of some aspects of the specification. This summary is not an extensive overview of the specification. It is intended to neither identify key or critical elements of the specification nor delineate any scope of the particular embodiments of the specification, or any scope of the claims. Its sole purpose is to present some concepts of the specification in a simplified form as a prelude to the more detailed description that is presented later.
Some embodiments of the invention are now described, by way of example only, and with reference to the accompanying drawings. The same reference number represents the same element or the same type of element on all drawings.
The figures and the following description illustrate specific exemplary embodiments. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described or shown herein, embody the principles of the embodiments and are included within the scope of the embodiments. Furthermore, any examples described herein are intended to aid in understanding the principles of the embodiments, and are to be construed as being without limitation to such specifically recited examples and conditions. As a result, the inventive concept(s) is not limited to the specific embodiments or examples described below, but by the claims and their equivalents.
In this embodiment, communication system 100 includes one or more computer networks 102-104. A computer network 102-104 is a group of computer systems and other computing hardware devices that are linked together through communication channels to facilitate communication and resource-sharing among a wide range of users. Computer network 102 may represent an open or unsecure global computer network providing a variety of information and communication facilities through interconnected networks using standardized communication protocols, one example of which is the Internet. Computer network 102 may include web servers 110, mail servers 111, database servers 112, data stores 113, and/or other elements that store documents and/or facilitate the transfer of the documents. Computer network 103 may represent a mobile communication network (also referred to as a mobile network or cellular network) configured to communicate with user terminals via wireless signals. Computer network 103 may include cellular towers 120, switches 121, gateways 122, servers 123 and/or other elements that store documents and/or facilitate the transfer of the documents. Computer network 104 may represent a secure computer network, such as a corporate network or enterprise network, providing a variety of information and communication facilities through interconnected networks. An enterprise network is a group of computers, servers, or other devices connected together in a building or in a particular area, which are all owned by the same company, entity, institution, etc. Computer network 104 may include file servers 130, mail servers 131, database servers 132, data stores 133, and/or other elements that store documents and/or facilitate the transfer of the documents.
Communication system 100 also includes a variety of user terminals 140-143, which may also be referred to as User Equipment (UE) or end user devices. User terminals 140-143 are hardware devices that are used directly by users 150 (i.e., end users) to access a service made available by a server or serving element of a computer network. For example, user terminal 140 may represent a desktop computer, user terminal 141 may represent a laptop computer, user terminal 142 may represent a mobile phone (e.g., a smartphone), and user terminal 143 may represent a personal digital assistant (PDA). Through user terminals 140-143, users 150 are able to access a variety of documents from one or more of computer networks 102-104. For example, users 150 may access email from a mail server 111/131, may access web pages from a web server 110, may access files from a file server 112/130, etc.
The amount of new documents available to users 150 may be massive, and users 150 may not realistically be able to view all of the new documents. In the embodiments described herein, a recommendation system parses the new documents and recommends a subset of the new documents to users 150. Instead of developing a static profile for the users 150 as with prior systems, the recommendation system as described herein is adaptive and is able to change with topics, trends, etc., in the new documents.
In this embodiment, recommendation system 200 includes a collector subsystem 202, which comprises circuitry, hardware, or means configured to accumulate, receive, or acquire information regarding documents that are circulated, disseminated, published, or otherwise distributed by a computer network. Collector subsystem 202 may query a variety of servers to acquire the information, may subscribe to updates of the information, may receive information pushed by servers, etc. In this embodiment, collector subsystem 202 is configured to identify historical information indicating consumption by users of past documents (e.g., older than one day, two weeks, one month, two months, etc.), and is configured to identify new or recent documents (e.g., published within the last day, within two weeks, within one month, within two months, etc.).
Recommendation system 200 further includes a vectorization subsystem 204, which comprises circuitry, hardware, or means configured to vectorize the documents. Vectorization or vectorizing refers to the parsing of the documents to extract constituents (e.g., terms or words of a text-based document), and assigning values to the constituents, such as to indicate a degree or frequency of the constituents in the documents. In this embodiment, vectorization subsystem 204 is configured to vectorize the historical information and the recent documents to generate matrices. For example, vectorization subsystem 204 is configured to generate a historical reading matrix based on the historical information, and a vector matrix for the recent documents and the past documents.
Recommendation system 200 further includes an estimator subsystem 206, which comprises circuitry, hardware, or means configured to generate an estimated reading matrix based on the historical reading matrix and the vector matrix.
Recommendation system 200 further includes a refiner subsystem 208, which comprises circuitry, hardware, or means configured to perform a factorization on the estimated reading matrix to generate a refined reading matrix. Matrix factorization (also referred to as matrix decomposition) refers to the decomposition of a matrix (e.g., V) into smaller matrices (e.g., W and H), and then reconstructing a refined matrix (e.g., V′) from the product of the smaller matrices (V′=WH).
Recommendation system 200 further includes a recommender subsystem 210, which comprises circuitry, hardware, or means configured to generate recommendations of subsets of the recent documents to the users based on the refined reading matrix. Recommender subsystem 210 may provide the recommendations to servers over a network, may present the recommendations directly to a user, such as through a Graphical User Interface (GUI), etc. A GUI manages the interaction between a computer system and a user through graphical elements, such as windows on a display.
One or more of the subsystems of recommendation system 200 may be implemented on a hardware platform comprised of analog and/or digital circuitry. One or more of the subsystems of recommendation system 200 may be implemented on a processor 220 that executes instructions stored in memory 222. Processor 220 comprises an integrated hardware circuit configured to execute instructions, and memory 222 is a computer readable storage medium for data, instructions, applications, etc., and is accessible by processor 220.
For method 300, it is assumed that documents from one or more computer networks are available to a pool of users. For example, news articles may be available for viewing by the users over the Internet, via an enterprise network, through email, etc. The users may read, view, or otherwise consume the documents based on their preferences, such as by viewing the news articles of interest to them. The documents described herein may be divided into past documents and recent documents. Recent documents (or new documents) are defined as documents published or otherwise available to users within a time period preceding the present date (e.g., within the last day, within two weeks, within one month, within two months, etc.), so that the documents are new to the user or new in the time period. Past documents are defined as documents published or otherwise available to users prior to the time period (e.g., older than one day, two weeks, one month, two months, etc.).
Collector subsystem 202 identifies recent documents that were published or otherwise available to users (step 302). For example, the recent documents may be news articles made available during the time period. Collector subsystem 202 identifies historical information indicating consumption (e.g., reading or viewing) of past documents by users (step 304). The historical information indicates reading patterns or reading interactions for the past documents by the users. The historical information may include user identities for the users, the past documents consumed by the users or metadata for the past documents, ratings or scores for the past documents provided by the users, etc. Unlike the past documents, the recent documents most likely do not have associated information indicating reading patterns by the users.
In
As indicated in
In
In
Estimator subsystem 206 may use a variety of techniques to calculate the estimated reading scores. In one embodiment, estimator subsystem 206 may use a cosine-averaged algorithm. One example of a cosine-averaged algorithm is as follows:
Equation 1 solves the estimated reading score rua of a user u for a recent document a∈An. Ap represents a set of past documents, and An represents a set of recent documents. The term r*ua′ denotes the actual reading score of a user u for a past document a′∈Ap. Cosine(a, a′) is the cosine of two row vectors corresponding to document a and a′ from vector matrix 412. The rationale is that if a and a′ are similar, their corresponding row vectors from vector matrix 412 will be similar and cosine(a, a′) will be close to 1. Thus, if r*ua′ is high, then a will contribute to a high value towards rua.
In another embodiment, estimator subsystem 206 may use different prediction or learning algorithms to calculate the estimated reading scores, such as Support Vector Machines (SVM), Deep Convolutional Neural Networks (CNN), etc.
In
In other embodiments, refiner subsystem 208 may use other matrix decomposition algorithms, such as Singular Value Decomposition (SVD), Lower-Upper (LU) decomposition, etc.
Recommender subsystem 210 then generates recommendations of the recent documents to the users based on refined reading matrix 430 (step 318). For example, recommender subsystem 210 may generate a list of recent documents for a user as a recommendation for the user. Recommender subsystem 210 may send the list of recent documents to a server in the network, which in turn allows the server to present the list to the user. Recommender subsystem 210 may alternatively present the list of recent documents to the user, such as through a GUI. For instance, if recommendation system 200 is implemented in an email client, then it may present the list to the user through the email client. If recommendation system 200 is implemented in a web server or a web browser, then it may present the list to the user through the web browser.
Method 300 may subsequently be repeated as desired. When method 300 repeats at a later time, newly-published documents will be available and defined as “recent documents”. Also, the time period used to define the recent documents and the past documents shifts to a later date. For example, assume for one iteration of method 300 that the present date is December 1, and the time period is set at one week. For this iteration, the documents published before November 24 (i.e., more than a week before December 1) are defined as past documents, and the documents published after November 24 are defined as recent documents. Assume for another iteration of method 300, that the present date is December 10, and the time period is again set at one week. For this iteration, the documents published before December 3 (i.e., more than a week before December 10) are defined as past documents, and the documents published after December 3 are defined as recent documents. Thus, newly-published documents are designated as “recent documents” over time, and some documents may transition from being designated as “recent documents” to “past documents”, which grows the corpus of past documents. Also, actual reading scores may be assigned to the documents that transition into “past documents” as they are consumed by the users. When method 300 is executed again at later dates, the refined reading matrix evolves based on the documents added to the corpus of past documents.
One benefit of recommendation system 200 is computational speed. In contrast to other machine learning algorithms, which easily take hours or days for running, recommendation system 200 performs calculations in a fraction of that time. This allows for real-time recommendations. Another benefit is that recommendation system 200 does not rely on a static profile of a user in making recommendations. Recommendation system 200 is adaptive in that the metrics for its computations change over time. Recommendation system 200 uses a sliding “present date” and time period to define what is a recent document and what is a past document. By doing so, the model used by recommendation system 200 changes based on how users consume the documents over time.
Any of the various elements or modules shown in the figures or described herein may be implemented as hardware, software, firmware, or some combination of these. For example, an element may be implemented as dedicated hardware. Dedicated hardware elements may be referred to as “processors”, “controllers”, or some similar terminology. When provided by a processor, the functions may be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which may be shared. Moreover, explicit use of the term “processor” or “controller” should not be construed to refer exclusively to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor (DSP) hardware, a network processor, application specific integrated circuit (ASIC) or other circuitry, field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), non-volatile storage, logic, or some other physical hardware component or module.
Also, an element may be implemented as instructions executable by a processor or a computer to perform the functions of the element. Some examples of instructions are software, program code, and firmware. The instructions are operational when executed by the processor to direct the processor to perform the functions of the element. The instructions may be stored on storage devices that are readable by the processor. Some examples of the storage devices are digital or solid-state memories, magnetic storage media such as a magnetic disks and magnetic tapes, hard drives, or optically readable digital data storage media.
As used in this application, the term “circuitry” may refer to one or more or all of the following:
(a) hardware-only circuit implementations (such as implementations in only analog and/or digital circuitry);
(b) combinations of hardware circuits and software, such as (as applicable):
(c) hardware circuit(s) and or processor(s), such as a microprocessor(s) or a portion of a microprocessor(s), that requires software (e.g., firmware) for operation, but the software may not be present when it is not needed for operation.
This definition of circuitry applies to all uses of this term in this application, including in any claims. As a further example, as used in this application, the term circuitry also covers an implementation of merely a hardware circuit or processor (or multiple processors) or portion of a hardware circuit or processor and its (or their) accompanying software and/or firmware. The term circuitry also covers, for example and if applicable to the particular claim element, a baseband integrated circuit or processor integrated circuit for a mobile device or a similar integrated circuit in server, a cellular network device, or other computing or network device.
Although specific embodiments were described herein, the scope of the disclosure is not limited to those specific embodiments. The scope of the disclosure is defined by the following claims and any equivalents thereof
