Patent Application
20190050432
The modern abundance of personal data from sources such as email, contacts, and documents cannot be overstated. Indeed, there exists a significant lack of and ever-growing need for even greater abilities to process such data in meaningful ways so as to provide a user with opportunities to do more than mere keyword searches or similar actions. Current systems offer limited use of information within personal and public data and generally provide a user with little more than typical search engine functionality.
There remains a need for a way to address these and other problems associated with the prior art. More particularly, there remains a need for greater leveraging of personal data for a user, particularly with regard to unstructured data.
The disclosed technology relates generally to data processing, query processing, and more particularly but not exclusively to systems and methods for processing document and text data. For example, knowledge may be harvested from unstructured data and subsequently relied on or used to provide a user with meaningful information that ties together multiple pieces of data from any of a number of personal data sources and, in some embodiments, public data sources.
The networked system 100 also includes three mobile electronic devices 108-112. Two of the mobile electronic devices, 108 and 110, are mobile communications devices such as cellular telephones or smart phones. The third mobile electronic device, 112, is a handheld device such as a personal data assistant (PDA) or tablet device.
The networked system 100 also includes a storage device 114, which may be a central database or repository, a local data store, or a remote storage device, for example. The storage device 114 may be accessible to any or all of the other devices 104-112, subject to limitations or restrictions by the devices 104-112, a third party, or the storage device 114 itself. The storage device 114 may be used to store some or all of the personal data that is accessed and/or used by any of the computers 104 and 106 or mobile electronic devices 108-112. In situations involving public data, the storage device 114 may also store any or all of the public data accessed and/or used by any of the computers 104 and 106 or mobile electronic devices 108-112.
The AOC 208 is part of the KELE 206, which includes various other subsystems such as an intent identification module 210, a learning module 212, a concept expansion module 214, a deep analysis and reasoning module 216, and various user data sources 218 that provide personal data and information. The AOC 208 is configured to interact with a knowledge store 220, such as the storage device 114 of
The system 300 also includes a knowledge store 306 configured to store knowledge, generally in the form of structured data. As used herein, the term structured data generally refers to data or information that is identifiable because it is organized in a structure. Structured data is typically searchable by data type within content, readily understood by computing devices, and efficiently organized for human readers. Structured data as described herein can generally be used to identify a person, place, or item involved with a particular field or industry, e.g., sales. Such structured data typically includes, but is not limited to, fields in a CRM application, such as contact information, account name, contact name, invoice number, and phone number.
Structured data is usually organized in such a way that it is readily and often easily searchable, presentable, or useable by an application or user. In contrast, the term unstructured data as used herein generally refers to data that has no identifiable structure. Unstructured data may include content that is similar or even identical to corresponding structured data but is not organized in such a way that it is readily or easily searchable, presentable, or useable by an application or user. Whereas data corresponding to a “sender” field in an email message is usually structured data, for example, the typical freeform text of the email body is generally unstructured data.
The data extractor and correlator 302 is configured to retrieve personal data from at least one of the personal data sources 304A-n. For example, the data extractor and correlator 302 may be configured to retrieve all incoming email messages subject to a filter, e.g., all email messages from a certain sender or originator. Alternatively or in addition thereto, the data extractor and correlator 302 may retrieve all documents created by or edited by the user. A functional or actual filter may be used to specify that only certain documents, e.g., documents pertaining to sales involving the user, are to be retrieved by the data extractor and correlator 302.
The data extractor and correlator 302 is further configured to extract information from unstructured data within the retrieved personal data. For example, an email message retrieved from the data extractor and correlator 302 may contain unstructured data such as freeform text in the subject or body of the message. In such a situation, the data extractor and correlator 302 may extract certain words, terms, or phrases, such as contact information or sales-related information, from the unstructured data within the message.
The data extractor and correlator 302 is further configured to correlate the extracted information with previously stored structured data, e.g., stored in the knowledge store 306, to generate additional structured data. For example, consider a situation in which the data extractor and correlator 302 extracts additional information, e.g., a secondary phone number extracted from the body of an email message, that pertains to a sales contact having information, e.g., a name and a primary phone number, that is already stored in the knowledge store 306. The extracted information (secondary phone number) will be correlated with the previously stored structured data (existing name and primary phone number) to generate additional structured data (secondary phone number added to or associated with the existing contact).
The knowledge store 306 is configured to store additional structured data as well as previously stored structured data. The data extractor and correlator 302 thus provides output in the form of enriched knowledge that may be stored within the storage device 306 and used in subsequent queries or applications by the user or other users or even other applications. For example, in the situation described above, a subsequent query by a user involving the sales contact may provide the secondary phone number without the user needing to perform an additional or more detailed search for the information.
Table 1 provides an example of different types of structured data that may be extracted from various types of personal data sources.
Table 2 provides an example illustrating how the data extractor and correlator 302 of
Certain embodiments of the system 300 of
A document harvester and indexer may be used to process and index documents including files, e.g., word processing files, spreadsheet files, presentation files, individual slides in presentation files, etc., calendar events, to do lists, notes, emails, email attachments, and web pages. These documents may be retrieved locally from a user's computer and/or remotely from network storage, e.g., a server that stores documents produced by a plurality of users, as well as from the Web, e.g., from web pages via Web application programming interfaces (APIs). The documents may also be tagged and/or clustered.
As documents are harvested, a word popularity dictionary may be created. Word popularity generally refers to a global dictionary containing high frequency words and weights. When a new document is harvested, for example, keywords that do not exist in the dictionary may be added. Stemming may be applied to obtain root words and text may be converted to lowercase. As a user interacts with the system by sending email, visiting web pages etc., the weights in the dictionary can be constantly updated. Keywords in frequently-accessed documents may be given higher weights while keywords in less-important documents may be given lower weights. Consequently, an up-to-date and accurate model of the user's universe and behavior may be effectively constructed.
In the example, the user profile module 408 is configured to interact with any number of user profiles 410A-n. Each user profile may correspond to one or more users. Also, any given user may be associated with multiple user profiles. For example, each user profile may correspond to a certain role, e.g., sales coordinator, that may be assigned to or associated with multiple users. Multiple user profiles 410A-n may correspond to a user's particular situation. For example, a user may have one user profile 410A for work-related items and a second user profile 410B for home-related items. Alternatively or in addition thereto, a user may have one or more profiles that correspond to activities with friends, one or more profiles that correspond to family activities, and one or more profiles that correspond to business-related events.
The user profile module 408 may interact with a knowledge store 406 such as the knowledge store 306 of
In certain embodiments, the user profile module 408 may interact with one or more public data sources 414. For example, a personal corpus or web data often do not provide enough information to build or update a user profile that is detailed or accurate enough for certain applications. In these embodiments, the user profile module 408 may proactively seek or passively receive public information pertaining to a contact whose information is stored by the knowledge store 406. If the new public information is different than the previously stored information, the user profile module 408 may direct the knowledge store 406 and/or one or more of the user profiles 410A-n to update the corresponding information accordingly.
The feedback module 516 may interact with one or both of the user profile module 508 and the knowledge store 506. In certain embodiments, the feedback module 516 may interact with one or more public data source 514 and may cause the user profile module 508 to alter or update one or more of the user profiles 510A-n based on interactions with the public data source(s) 514. In certain embodiments, the feedback module 516 may interact directly with a user associated with one of the user profiles 510A-n. Alternatively or in addition thereto, the feedback module 516 may interact directly with one or more user applications 512A-n, such as the user applications 412A-n of
Consider a situation in which user profile 510B involves a particular sales contact whose contact information just changed and is broadcast via the public data source 514. The feedback module 516 may direct the user profile module 508 to update one or more of the user profiles 510A-n with the new public information concerning the sales contact. The user profiles 510A-n can be continually updated and enriched as more searches are conducted and in an increasingly refined manner. For example, suggestions provided to a user based on his or her user profile(s) may be increasingly relevant as time goes on.
In embodiments where the feedback module 516 interacts with one or more user applications 512A-n, the feedback module 516 may be triggered to direct the user profile module 508 to update one or more of the user profiles 510A-n responsive to the interaction with the user application(s) 512A-n. For example, if the feedback module 516 detects a user updating a contact mailing address in user application 512B, the feedback module 516 may direct the user profile module 508 to update any of the user profiles 510A-n that include a mailing address for the contact.
At 604, information is extracted from unstructured data within the data retrieved at 602. For example, a data extractor and correlator, such as the data extractor and correlator 302 of
The information extraction performed at 604 may be accomplished by breaking at least one sentence into subject, verb, and object (SVO), extracting phrases that link a subject to an object, extracting at least one word in close proximity to an identified feature or service, extracting at least one word in close proximity to a known quality, or any combination thereof. Features with certain quality or derived quality ratings may be tagged for reviews, for example. Also, structures that approximate concepts from documents with and without prior semantic understanding may be constructed.
At 606, some or all of the extracted information is correlated with previously stored structured data to generate additional structured data. For example, a data extractor and correlator, such as the data extractor and correlator 302 of
In certain embodiments, the retrieved data includes supplemental structured data, e.g., structured data that has not yet been stored within a knowledge store. In these situations, the data extractor and correlator may correlate the supplemental structured data with one or both of the previously stored structured data and the additional structured data to generate further structured data that may be stored by the knowledge store.
Certain embodiments may include retrieving public data from one or more public data sources. In these embodiments, a data extractor and correlator may extract public information from unstructured data within the retrieved public data and correlate the extracted public information with previously stored structured data to generate further additional structured data that may be stored by the knowledge store.
In certain embodiments, a user profile, such as the user profiles 410A-n of
At 614, a natural language query is received from a user. For example, a user wishing to research a particular sales lead may provide the following query: “has there been any recent progress with sales lead XYZ Manufacturing?” The system then generates a response to the natural language query received at 614, as indicated at 616. The response is based at least in part on one or both of the previously stored data and the additional structured data. For example, if the stored structured data contains information pertaining to XYZ manufacturing, the generated response may provide said information to the user in the response.
At 704, semantic analysis is performed on the raw content. For example, a semantic analysis module may be configured to determine semantic information based on unstructured data within the retrieved data. A data extractor and correlator, such as the data extractor and correlator 302 of
Certain embodiments include performing a search of one or more data sources based on results of the semantic analysis performed at 704. Such embodiments may include performing a search of one or more personal data sources, as indicated by 706, or performing a search of one or more public data sources, as indicated by 708.
At 710, an additional analysis is performed based at least in part on the results of the search performed at either 706 or 708. In certain embodiments, a user profile, such as the user profiles 410A-n of
In the example, the data sources 804A-E collectively yield five pieces of structured data 805A that may be retrieved, for example, by a data extractor and correlator: sales contact information (name and email address), account name, contact name, invoice number, and phone number. The data sources 804A-E also provide various pieces of unstructured data 805B: two proper names (person and company), meeting time, two invoice numbers, a phone number, meeting-specific information, and sales-specific information.
Certain information, e.g., invoice numbers, from the unstructured data 805B may be correlated with the structured data 805A. Such correlation may include identifying, extracting, or building at least one relationship between the extracted information and previously stored structured data. For example, one or more features identified within the extracted information may be tagged or otherwise marked for subsequent operations. Parts of speech analysis may also be performed and then enriched by relationship determinations.
In certain embodiments, correlation and relationship building may include categorizing one or more portions of the unstructured data. Portions of the unstructured data 805B may each be broken into subject, verb, and object (SVO). Phrases linking a subject to an object may be extracted. A determination may be made as to certain words in close proximity to an identified feature or service, known quality, or any combination thereof.
Consider an example in which the word “Apple” in unstructured free-flowing data could have multiple meanings. The word could refer to a fruit, the name of a company, the name of a person, etc. Relationships may be established to decipher the meaning of certain unstructured data. A possible association to each of these entities can be created and those that have a high probability based on parts of speech analysis and entity relationship identification, for example, may be strengthened. In the present example, an inference may be made that “Apple” is a company based on its learning from a prior structured data analysis. Table 3 provides an indication as to the relationships for “Apple” that may be identified from structured data. In the example, the word “Apple” occurs along with “Apple Thomas” and “Rob Cramer.”
In certain embodiments, one or more patterns may be extracted or determined from structured data, such as previously stored structured data, to create pattern fingerprints. Patterns may subsequently be extracted from the unstructured data using these pattern fingerprints. For example, structured data may be used to construct a pattern fingerprint knowledge base and then use the pattern fingerprinting knowledge to recognize similar items from unstructured data and establish their relationship with various entities. For example, fingerprint data patterns can be learned and determined for sales and/or contact-related attributes such invoice number, P.O. number, and phone number. These learnings may be applied to identify similar patterns in unstructured data and identify additional relationships between entities.
Certain implementations of the disclosed technology may include personalized searching capabilities and features, personalized content delivery, personalized advertisement delivery, intelligence gathering and analysis, and automated augmentation of knowledge bases.
Embodiments of the disclosed technology may be implemented as machine-directed methods or physical devices. Accordingly, certain implementations may take the form of an entirely-hardware embodiment, an entirely-software embodiment, or an embodiment combining both hardware and software aspects. For example, some or all of the components for any given embodiment may be computer-implemented components.
Having described and illustrated the principles of the invention with reference to illustrated embodiments, it will be recognized that the illustrated embodiments may be modified in arrangement and detail without departing from such principles, and may be combined in any desired manner. And although the foregoing discussion has focused on particular embodiments, other configurations are contemplated. In particular, even though expressions such as “according to an embodiment of the invention” or the like are used herein, these phrases are meant to generally reference embodiment possibilities, and are not intended to limit the invention to particular embodiment configurations. As used herein, these terms may reference the same or different embodiments that are combinable into other embodiments.
Consequently, in view of the wide variety of permutations to the embodiments described herein, this detailed description and accompanying material is intended to be illustrative only, and should not be taken as limiting the scope of the invention. What is claimed as the invention, therefore, is all such modifications as may come within the scope and spirit of the following claims and equivalents thereto.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 13287983 | Nov 2011 | US |
| Child | 16165242 | US |
