System and method for analysis of an opinion expressed in documents with regard to a particular topic

• US 20100023311A1
• Filed: 06/08/2007
• Published: 01/28/2010
• Est. Priority Date: 09/13/2006
• Status: Active Grant

• Current Assignee: University of Maryland, College Park
• Original Assignee: University of Maryland, College Park
• Inventors: Subrahmanian, Venkatramanan Siva, Dorr, Bonnie J., Cesarano, Carmine, Picariello, Antonio, Reforgiato, Diego Recupero, Sagoff, Amelia
• Granted Patent: US 8,296,168 B2
• Time in Patent Office: Days
• US Class Current: 704/2
• CPC Class Codes: G06F 16/345 Summarisation for human users
• CPC Class Codes: G06F 40/263 Language identification
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

System and method for analysis of an opinion expressed in documents on a particular topic computes opinion strength on a continuous numeric scale, or qualitatively. A variety of opinion scoring techniques are plugged in to score opinion expressing words and sentences in documents. These scores are aggregated to measure the opinion intensity of documents. Multilingual opinion analysis is supported by capability to concurrently identify and visualize the opinion intensity expressed in documents in multiple languages. A multi-dimensional representation of the measured opinion intensity is generated which is agreeable with multi-lingual domain.

Claims

1. A system for analysis of an opinion expressed in documents with regard to a particular topic, comprising:

a data processing system including an opinion analysis module;

a specification unit specifying at least one source of documents, and a topic of interest;

a document retrieving unit operationally coupled to said specification unit and retrieving at least one document d on said topic of interest from said at least one source of documents for entrance into said opinion analysis module;

said opinion analysis module having an opinion intensity scoring unit computing, at least on a substantially continuous numerical scale, an intensity of an opinion expressed in said at least one document on said topic of interest; and

an opinion intensity representation unit outputting said computed opinion intensity in a multi-dimensional format.

2. The system of claim 1, wherein said opinion intensity scoring unit includes a quantitative opinion analysis module computing said opinion intensity on said substantially continuous numerical scale.

3. The system of claim 2, wherein said quantitative opinion analysis module further computes said opinion intensity on a binary scale.

4. The system of claim 1, wherein said opinion intensity scoring unit includes a qualitative opinion analysis module qualitatively rating intensity of opinion expressed in said at least one document on said topic of interest.

5. The system of claim 1, wherein said specification unit further specifies a time period of interest.

6. The system of claim 1, further comprising:

a translation module operatively coupled between said document retrieving unit and said opinion analysis module, said translation module including a language identifying unit identifying a language used in said at least one document d.

7. The system of claim 6, wherein said translation module further includes at least one translation unit translating said identified language used in said at least one document d into English.

8. The system of claim 1, wherein said opinion intensity representation unit including a multi-dimensional visualization engine operatively coupled to said opinion analysis module and outputting said computed opinion intensity of a plurality of documents d using multiple text languages.

9. The system of claim 1, further comprising a Scored Opinion Expressing Words Bank operatively coupled to said opinion analysis module and including pre-calculated word scoring functions assigning respective scores to predetermined opinion expressing words and phrases.

10. The system of claim 1, wherein said opinion intensity scoring unit further comprises a sentence scoring unit, said opinion intensity scoring unit calculating said intensity of an opinion expressed in said at least one document d by aggregating opinion intensity scores of relevant sentences in said at least one document d calculated in said sentence scoring unit.

11. The system of claim 10, wherein said sentences scoring unit further includes a words scoring unit assigning opinion intensity scores to opinion expressing words and synonyms thereof found in relevant sentences.

12. The system of claim 1, wherein said opinion intensity scoring unit further includes a Hybrid Evaluation Module assigning scores to said opinion intensity of said at least one document by merging a plurality of opinion intensity scoring techniques.

13. A system for analysis of an opinion expressed in documents on a particular topic in a multi-lingual domain, comprising:

a data processing system including;

a data retrieving unit retrieving documents relevant to a topic of interest from a plurality of sources of documents in a multi-lingual domain;

a translation module operatively coupled to said data retrieving unit and receiving said retrieved documents therefrom, said translation module including at least one translator unit and a language identifying unit identifying a language used in each retrieved document received at said translation module; and

an opinion analysis module operatively coupled to said translation module to compute intensity of opinions on said topic of interest expressed in documents received from said translation module, said opinion analysis module including an opinion analysis unit adapted to at least one respective language;

wherein, if said identified language corresponds to said at least one respective language, said translation module outputs said each retrieved document to said opinion analysis module, and

if said identified language differs from said at least one respective language, said translation module applies said at least one translator unit to said each retrieved document to translate said identified language into said at least one respective language; and

wherein upon said translation has been accomplished, said translation module outputs translated said each retrieved document to said opinion analysis module.

14. The system of claim 13, further comprising a multi-dimensional visualization engine operatively coupled to said opinion analysis module to output said calculated opinion intensity in a multi-dimensional format agreeable with a multi-lingual domain of the retrieved documents.

15. The system of claim 14, further comprising an opinion database unit coupled to said opinion analysis module and said multi-dimensional visualization engine, said opinion database unit including a database containing identification information of each retrieved document, topic extracted from said each retrieved document, relevant metadata, and opinion scores of said each retrieved documents.

16. The system of claim 15, further comprising an opinion cache unit operatively coupled to said opinion database unit and said multi-dimensional visualization engine, said opinion cache unit operatively coupled to said opinion database unit and said multi-dimensional visualization engine, said opinion cache unit including a high-speed in-memory server containing precomputed data.

17. The system of claim 13, wherein said opinion analysis module further includes an opinion intensity scoring unit assigning scores to an intensity of an opinion expressed in documents on said topic of interest on substantially continuous and binary numerical scale, and rating the same qualitatively.

18. A method for analysis of an opinion expressed in documents on a particular topic of interest, comprising the steps of:

specifying sources of documents, topic of interest, and time period of interest to a data processing system including an opinion analysis module adapted for analyzing documents in at least one respective language;

retrieving documents D on said topic of interest for said time period from said sources of documents;

identifying a language of each document d of said retrieved documents D;

entering said each document d into said opinion analysis module if said identified language of said each document d corresponds to said at least one respective language,

otherwise, translating said each document d from said identified language into said at least one respective language, and sending a translated said each document d to said opinion analysis module;

computing an intensity of an opinion expressed in said each document d on said topic of interest in said opinion analysis module; and

outputting said computed intensity of opinion in a multi-dimensional format.

19. The method of claim 18, wherein said intensity of opinion is computed on a substantially continuous numerical scale.

20. The method of claim 18, wherein said intensity of opinion is computed on a binary numerical scale.

21. The method of claim 18, wherein said intensity of opinion is computed qualitatively.

22. The method of claim 18, comprising the steps of:

prior to computing said intensity of opinion, forming a Scored Opinion Expressing Words Bank containing precalculated word scoring functions assigning respective scores to predetermined opinion expressing words and phrases.

23. The method of claim 22, wherein formation of the Scored Opinion Expressing Words Bank further comprises the steps of:

selecting a collection D_test of training documents,

scoring opinion intensity of each document in said collection D_test by a plurality of human individuals,

for each document from said collection D_test, forming a multiset of said opinion intensities scored by said plurality of human individuals, and

scoring opinion expressing words and phrases.

24. The method of claim 22, wherein said computation of the intensity of the opinion expressed in said each document d on said topic of interest further comprises the steps of:

extracting a set of relevant sentences from said each document d,

in each said relevant sentence s, finding opinion expressing words ω and

synonyms ω

′

thereof,

for each of said ω and

ω

′

, extracting a corresponding score from said Scored Opinion Expressing Words Bank,

adding said extracted scores for all ω and

ω

′

in said each relevant sentence s, and

averaging said added score of all relevant sentences s in said set thereof in said each document d.

25. The method of claim 22, wherein said computation of the intensity of the opinion expressed in said each document d on said topic of interest further comprises the steps of:

assigning “

initial”

score sc(s) and sc(s_n) to each sentence s and s_n expressing opinion and not expressing opinion, respectively, on said topic of interest,

splitting said scored sentences s and s_n,

for each s and each s_n, finding a distance therebetween in said each document d,

for each said s, calculating an impact of all s_n thereon, and

averaging all sc(s) and impacts of all sc(s_n).

26. The method of claim 22, wherein said computation of the intensity of the opinion expressed in said each document d on said topic of interest further comprises the steps of:

establishing a list of lexographic templates,

applying said lexographic templates to the sentences in said each document d, extracting relevant sentences in said each document matching at least one of said lexographic templates, and

scoring the relevant sentences.

27. The method of claim 22, wherein said computation of the intensity of the opinion expressed in said each document d on said topic of interest further comprises the steps of:

specifying a document collection D,

specifying a document collection D_test,

associating with each document dε

D a first scoring vector corresponding to a plurality of scoring techniques,

scoring each said document dε

D in correspondence with said first scoring vector,

associating with each document d_iε

D_test a second scoring vector corresponding to a plurality of scoring techniques,

scoring each said document d_iε

D_test in correspondence with said second scoring vector,

scoring each said document d_iε

D_test by human individuals,

for each scored dε

D, finding r-nearest neighboring documents d_i(i=1, . . . , r) scored by the second scoring vector,

finding scores assigned by human individuals to said documents d_i (I=1, . . . , r), and

averaging said scores assigned to said documents d_i(i=1, . . . , r) by human individuals.

28. The method of claim 21, wherein said step of qualitative computation of the intensity of opinion comprises the steps of:

specifying a document collection D_test,

establishing a rating scale including a list of words positioned in a predetermined order of positiveness of an opinion,

defining threshold ranges for each word on said rating scale by applying scoring of documents in said document collection D_test by human individuals,

scoring each said document d quantitatively, and

assigning a qualitative rate to each said document d in accordance with said threshold ranges.

29. The method of claim 18, wherein said identification of the language used in said each document d further comprising the steps of:

specifying list of countries,

for each country, specifying a list of languages Ai,

extracting “

n”

random words from “

n”

random documents from said documents collection D,

applying a translator “

Ai English”

to said “

n”

random words,

calculating a number N_i of words translated into English,

if said N_i exceeds a predetermined threshold percentage, declaring the language A_i for said documents D.