System and method for sentiment-based text classification and relevancy ranking

• US 8,166,032 B2
• Filed: 04/09/2009
• Issued: 04/24/2012
• Est. Priority Date: 04/09/2009
• Status: Active Grant

• Current Assignee: MarketChorus, Inc.
• Original Assignee: MarketChorus, Inc.
• Inventors: Sommer, Matthew S., Goldstein, Lawrence N.
• Primary Examiner(s): Hoang, Son T
• Application Number:
  US12/384,880
• Publication Number:
  US 20100262454A1
• Time in Patent Office: 1,111 Days
• Field of Search: 707/736, 707/758
• US Class Current: 707/736
• CPC Class Codes: G06F 16/355 Class or cluster creation or modification
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising

Abstract

The sentimental significance of a group of historical documents related to a topic is assessed with respect to change in an extrinsic metric for the topic. A unique sentiment binding label is included to the content of actions documents that are determined to have sentimental significance and the group of documents is inserted into a historical document sentiment vector space for the topic. Action areas in the vector space are defined from the locations of action documents and singular sentiment vector may be created that describes the cumulative action area. Newly published documents are sentiment-scored by semantically comparing them to documents in the space and/or to the singular sentiment vector. The sentiment scores for the newly published documents are supplemented by human sentiment assessment of the documents and a sentiment time decay factor is applied to the supplemented sentiment score of each newly published documents. User queries are received and a set of sentiment-ranked documents is returned with the highest age-adjusted sentiment scores.

Claims

1. A computer implemented method of assessing human sentiment from a group of documents, each document in the group of documents having a plurality of terms and being digitally represented in a computer, the method comprising:

receiving, by the computer, a group of documents, each document in the group of documents comprising a context of a plurality of terms and all documents in the group of documents representative of a particular topic;

constructing, by the computer, a document sentiment vector space from the group of documents, wherein construction of the document sentiment vector space comprises;

assessing sentimentality of each document in the group of documents toward the topic, by the computer, wherein sentimentality represents human emotion toward the topic, comprising;

deriving a publication date for each document in the group of documents;

electing an extrinsic metric for the particular topic for assessing the sentimentality toward the topic, the extrinsic metric being related to an affirmative and intentional human action with a value of the extrinsic metric being indicative of the human action;

receiving extrinsic metric historical data for each document in the group of documents proximate to the respective publication date for each document; and

examining the extrinsic metric historical data for each document proximate to the respective publication date for each document over a timeframe of influence for changes in the value of the extrinsic metric, wherein the timeframe of influence is a predetermined time period in which a context of a document influences humans to undertake an affirmative and intentional human action resulting in a change in the value of the extrinsic metric;

identifying sentimentally significant documents in the group of documents with heightened sentimentality toward the particular topic, by the computer, comprising;

receiving a sentiment value for the change in the extrinsic metric historical data, the sentiment value being indicative of sentimental significance; and

comparing the sentiment value to the changes in the value of the extrinsic metric over the timeframe of influence for each document in the group;

labeling the identified sentimentally significant documents, in the computer, by including a unique sentiment binding term in the context of the plurality of terms;

representing, by the computer, each document in the group of documents in the document sentiment vector space;

defining, by the computer, a region of sentimental significance in the document sentiment vector space based on an occurrence of document representations for the identified sentimentally significant documents with the unique sentiment binding term;

receiving, by the computer, a query string; and

assessing, by the computer, the sentimentality of the query string by comparing a representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space.

2. The method of claim 1, wherein the particular topic is one of a person, occupation, athlete, entertainer, politician, humanitarian, political party, product, film, equity, cause, finance, complaint, patent, sport'"'"'s team and country.

3. The method of claim 1, wherein the particular topic is an equity and the extrinsic metric is one of selling price, change in the selling price, ratio of the change in the selling price to an average price, sales volume, change in the sales volume and ratio of the change in the sales volume to an average volume.

4. The method of claim 1, wherein constructing, by the computer, the document sentiment vector space from the group of document further comprises:

creating a term-by-document matrix of all terms occurring in the group of documents; and

decomposing the term-by-document matrix into a term-by-concepts matrix and a documents-by-concepts matrix.

5. The method of claim 4, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a singular sentiment vector from a product of document vectors from the documents-by-concepts matrix representing the identified sentimentally significant documents.

6. The method of claim 4, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a binding term vector from term vector in the term-by-concepts matrix representing the unique sentiment binding term.

7. The method of claim 4, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating an action document vector for each document vector from the documents-by-concepts matrix representing a sentimentally significant document.

8. The method of claim 7, wherein assessing the sentimentality of the query string, by the computer, by comparing the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

selecting an action document vector most semantically similar to the query string for assessing the sentimentality of the query string.

9. The method of claim 1, wherein the query string is a term and assessing the sentimentality of the query string by comparing, by the computer, the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

selecting a query term vector from a term-by-concepts matrix for the query string as the representation of the query string; and

comparing the query term vector for semantic similarity to the region of sentimental significance in the document sentiment vector space.

10. The method of claim 1, wherein the query string is a plurality of terms and assessing the sentimentality of the query string by comparing, by the computer, the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

creating a query term vector as the representation of the query string for the query string from a vector sum of query term vectors from a term-by-concepts matrix, for each of the plurality of terms occurring in the query string; and

comparing the query term vector for semantic similarity to the region of sentimental significance in the document sentiment vector space.

11. The method of claim 1, wherein the query string is a document having a document publication date and comprises of a plurality of document terms and assessing the sentimentality of the a query string, by the computer, by comparing a representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

creating a query term vector as the representation of the query string for the query string from a vector sum of query term vectors from a term-by-concepts matrix, for each of the plurality of terms co-occurring in the query string and the term-by concepts matrix; and

comparing the query term vector for semantic similarity to the region of sentimental significance in the document sentiment vector space.

12. The method of claim 11, further comprises:

quantizing, by the computer, a sentiment score for the sentimental significance of the query string to the particular topic based on the semantic similarity of the representation of the query string to the region of sentimental significance in the document sentiment vector space, wherein the sentiment score being indicative of the sentimental significance of a contemporaneous publication toward the particular topic.

13. The method of claim 12, wherein the document is a contemporaneous publication representative of the particular topic, the contemporaneous publication having a contemporaneous publication date and comprising a context of a plurality of publication terms, wherein the sentiment score being indicative of the sentimental significance of the contemporaneous publication toward the particular topic.

14. The method of claim 13, further comprises:

receiving, by the computer, a second query string comprising a second contemporaneous publication representative of the particular topic, the second contemporaneous publication having a second contemporaneous publication date and comprising a second context of a second plurality of publication terms;

assessing the sentimentality of the second contemporaneous publication, by the computer, by comparing a representation of the second query string for semantic similarity to the region of sentimental significance in the document sentiment vector space; and

quantizing, by the computer, a second sentiment score for the sentimental significance of the second query string to the particular topic based on the semantic similarity of the representation of the second query string to the region of sentimental significance in the document sentiment vector space, wherein the second sentiment score being indicative of the sentimental significance of the second contemporaneous publication toward the particular topic.

15. The method of claim 14, further comprises:

sentiment ranking, by the computer, the contemporaneous publication and the second contemporaneous publication based on their respective sentiment score and second sentiment score.

16. The method of claim 15, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a singular sentiment vector from a product of document vectors from the documents-by-concepts matrix representing the identified sentimentally significant documents.

17. The method of claim 15, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating the binding term vector from term vector in the term-by-concepts matrix representing the unique sentiment binding term.

18. The method of claim 15, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating an action document vector for each document vector from the documents-by-concepts matrix representing a sentimentally significant document.

19. The method of claim 18, wherein assessing the sentimentality of the query string, by the computer, by comparing the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

selecting an action document vector most semantically similar to the query string for assessing the sentimentality of the query string.

20. The method of claim 12, further comprises:

receiving, by the computer, a sentiment time decay factor for publications, wherein the sentiment time decay factor reflects a limited timeframe of influence of publications on the extrinsic metric relative to the publication dates for the publications;

finding, by the computer, an age sentiment score for the on temporaneous publication by applying the sentiment time decay factor to the sentiment score for age relative to the contemporaneous publication date;

finding, by the computer, a second age sentiment score for a second contemporaneous publication by applying the sentiment time decay factor to the second sentiment score for age relative to the second contemporaneous publication date; and

sentiment ranking, by the computer, the contemporaneous publication and the second contemporaneous publication based on their respective age sentiment score and second age sentiment score.

21. The method of claim 20, wherein defining the region of the sentimental significance in the document sentiment vector space further comprises:

creating a singular sentiment vector from a product of document vectors from the documents-by-concepts matrix representing the identified sentimentally significant documents.

22. The method of claim 20, wherein defining the region of the sentimental significance in the document sentiment vector space further comprises:

creating a binding term vector from term vector in the term-by-concepts matrix representing the unique sentiment binding term.

23. The method of claim 20, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating an action document vector for each document vector from the documents-by-concepts matrix representing a sentimentally significant document.

24. The method of claim 23, wherein assessing the sentimentality of the query string, by the computer, by comparing the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

selecting an action document vector most semantically similar to the query string for assessing the sentimentality of the query string.

25. The method of claim 14, further comprises:

receiving, by the computer, a trusted group sentiment score for the contemporaneous publication, the trusted group sentiment score being indicative of the sentiment of a trusted group of humans toward the particular topic with respect to the contemporaneous publication;

receiving, by the computer, a second trusted group sentiment score for the second contemporaneous publication, the second trusted group sentiment score being indicative of the sentiment of the trusted group of humans toward the particular topic with respect to the second contemporaneous publication;

finding, by the computer, a cumulative sentiment score for the contemporaneous publication from the sentiment score and the trusted group sentiment score; and

finding, by the computer, a second cumulative sentiment score for the second contemporaneous publication from the sentiment score and the trusted group sentiment score;

receiving, by the computer, a sentiment time decay factor for publications, wherein the sentiment time decay factor reflects a limited timeframe of influence of the publications on the extrinsic metric relative to the publication dates for the publications;

finding, by the computer, an age sentiment score for the contemporaneous publication by applying the sentiment time decay factor to the cumulative sentiment score for age relative to the contemporaneous publication date;

finding, by the computer, a second age sentiment score for the second contemporaneous publication by applying the sentiment time decay factor to the second cumulative sentiment score for age relative to the second contemporaneous publication date; and

sentiment ranking, by the computer, the contemporaneous publication and the second contemporaneous publication based on their respective age sentiment score and second age sentiment score.

26. The method of claim 14, further comprises:

receiving, by the computer, a global community sentiment score for the contemporaneous publication, the global community sentiment score being indicative of the sentiment of a global community of humans toward the particular topic with respect to the contemporaneous publication;

receiving, by the computer, a second global community sentiment score for the second contemporaneous publication, the second global community sentiment score being indicative of the sentiment of the global community of humans toward the particular topic with respect to the second contemporaneous publication;

receiving, by the computer, a trusted group sentiment score for the contemporaneous publication, the trusted group sentiment score being indicative of the sentiment of a trusted group of humans toward the particular topic with respect to the contemporaneous publication;

receiving, by the computer, a second trusted group sentiment score for the second contemporaneous publication, the second trusted group sentiment score being indicative of the sentiment of the trusted group of humans toward the particular topic with respect to the second contemporaneous publication;

finding, by the computer, a cumulative sentiment score for the contemporaneous publication relative to the sentiment score, the trusted group sentiment score and the global community sentiment score; and

finding, by the computer, a second cumulative sentiment score for the second contemporaneous publication relative to the second sentiment score, the second trusted group sentiment score, and the second global community sentiment score;

receiving, by the computer, a sentiment time decay factor for publications, wherein the sentiment time decay factor reflects a limited timeframe of influence of the publications on the extrinsic metric relative to the publication dates for the publications;

finding, by the computer, an age sentiment score for the contemporaneous publication by applying the sentiment time decay factor to the cumulative sentiment score for age relative to the contemporaneous publication date;

finding, by the computer, a second age sentiment score for the second contemporaneous publication by applying the sentiment time decay factor to the second cumulative sentiment score for age relative to the second contemporaneous publication date; and

sentiment ranking, by the computer, the contemporaneous publication and the second contemporaneous publication based on their respective age sentiment score and second age sentiment score.

27. The method of claim 26, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a singular sentiment vector from a product of document vectors from the documents-by-concepts matrix representing the identified sentimentally significant documents.

28. The method of claim 26, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a binding term vector from term vector in the term-by-concepts matrix representing the unique sentiment binding term.

29. The method of claim 27, wherein defining the region of sentimental significance in the document sentiment vector space further comprises:

creating a binding term vector from term vector in the term-by-concepts matrix representing the unique sentiment binding term.

30. The method of claim 29, wherein assessing the sentimentality of the query string, by the computer, by comparing the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

selecting an action document vector most semantically similar to the query string for assessing the sentimentality of the query string.

31. The method of claim 3, wherein assessing the sentimentality of the query string, by the computer, by comparing the representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space further comprises:

finding a dot product of the representation of the query string and the region of sentimental significance.

32. The method of claim 1, wherein deriving the publication date for each document in the group of documents further comprises finding a explicit publication date associated with each document in the group, inferring the publication date, implying the publication date, and assuming the publication date.

33. A computer implemented method of assessing human sentiment from a group of documents, each document in the group of documents having a plurality of terms and being digitally represented in a computer, the method comprising:

receiving, by the computer, a group of documents, each document in the group of documents comprising a context of a plurality of terms and all documents in the group of documents representative of a particular topic;

comprising;

receiving a sentiment term meaning model for defining sentimental meanings of terms; and

sentiment scoring the sentiment of each document in the group of documents based on the sentiment term meaning model and the plurality of terms in the respective document in the group of documents;

constructing, by the computer, a document sentiment vector space from the group of documents, wherein construction of the document sentiment vector space comprises;

assessing sentimentality of each document in the group of documents toward the topic, comprising;

deriving a publication date for each document in the group of documents;

electing an extrinsic metric for the particular topic for assessing the sentimentality toward the topic, the extrinsic metric being related to an affirmative and intentional human action with a value of the extrinsic metric being indicative of the human action;

receiving extrinsic metric historical data for each document in the group of documents proximate to the respective publication date for each document; and

examining the extrinsic metric historical data for each document proximate to the respective publication date for each document over a timeframe of influence for changes in the value of the extrinsic metric, wherein the timeframe of influence is a predetermined time period in which a context of a document influences humans to undertake an affirmative and intentional human action resulting in a change in the value of the extrinsic metric;

identifying, by the computer, sentimentally significant documents in the group of documents with heightened sentimentality toward the particular topic, wherein sentimentality represents human emotion toward the topic, comprising;

receiving an action sentiment score indicative of sentimental significance;

comparing the sentiment score of each document in the group of documents to the action sentiment score;

receiving a sentiment value for the change in the extrinsic metric historical data, the sentiment value being indicative of sentimental significance; and

comparing the sentiment value to the changes in the value of the extrinsic metric over the timeframe of influence for each document in the group;

labeling, by the computer, the identified sentimentally significant documents by including a unique sentiment binding term in the context of the plurality of terms;

representing, by the computer, each document in the group of documents in the document sentiment vector space;

defining, by the computer, a region of sentimental significance in the document sentiment vector space based on an occurrence of document representations for the identified sentimentally significant documents with the unique sentiment binding term;

receiving, by the computer, a query string; and

assessing, by the computer, the sentimentality of the query string by comparing a representation of the query string for semantic similarity to the region of sentimental significance in the document sentiment vector space.