Techniques for understanding the aboutness of text based on semantic analysis

• US 9,632,999 B2
• Filed: 04/03/2015
• Issued: 04/25/2017
• Est. Priority Date: 04/03/2015
• Status: Active Grant

• Current Assignee: Klangoo
• Original Assignee: Klangoo
• Inventors: Azzi, Johnny, Issa, Romeo, Saba, Walid, Touma, Eddy
• Primary Examiner(s): Baker, Charlotte M
• Application Number:
  US14/678,901
• Publication Number:
  US 20160292145A1
• Time in Patent Office: 753 Days
• Field of Search: 704 9, 704 10, 706 14, 706 54, 345467
• CPC Class Codes: G06F 40/211 Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
• CPC Class Codes: G06F 40/30 Semantic analysis

Abstract

In one embodiment of the present invention, a semantic analyzer translates a text segment into a structured representation that conveys the meaning of the text segment. Notably, the semantic analyzer leverages a semantic network to perform word sense disambiguation operations that map text words included in the text segment into concepts—word senses with a single, specific meaning—that are interconnected with relevance ratings. A topic generator then creates topics on-the-fly that includes one or more mapped concepts that are related within the context of the text segment. In this fashion, the topic generator tailors the semantic network to the text segment. A topic analyzer processes this tailored semantic network, generating a relevance-ranked list of topics as a meaningful proxy for the text segment. Advantageously, operating at the level of concepts and topics reduces the misinterpretations attributable to key word and statistical analysis methods.

Claims

1. A computer-implemented method for interpreting text segments based on word sense, the method comprising:

parsing a text segment to generate one or more text-based words and related syntactic information;

mapping, via a processor, each of the one or more text-based words to at least one concept included in a database and based on a semantic network that includes the at least one concept and one or more relevance ratings associated with the at least one concept, wherein each concept included in the semantic network is associated with a meaning and at least one word;

generating a plurality of topics based on the mappings and the syntactic information, wherein each topic includes one or more of the concepts included in the semantic network;

for each topic included in the plurality of topics, calculating a topic relevance rating between the topic and at least another topic included in the plurality of topics based on the relevance ratings between the one or more concepts included in the topic and one or more concepts included in the another topic;

ranking the plurality of topics based on the topic relevance ratings to generate a ranked topic list; and

outputting an element for display based on the ranked topic list.

2. The computer-implemented method of claim 1, wherein the semantic network includes a first concept and a second concept, the first concept is associated with a first word that is defined in a dictionary, and the second concept is associated with a second word that is a proper noun and is not defined in the dictionary.

3. The computer-implemented method of claim 1, wherein mapping each of the one or more text-based words comprises performing one or more word sense disambiguation operations.

4. The computer-implemented method of claim 3, wherein the one or more word sense disambiguation operations for a first text-based word included in the one or more text-based words comprises:

compositing a first meaning score for the first text-based word based on a first concept included in the at least one concept;

compositing a second meaning score for the first text-based word based on a second concept included in the at least one concept;

performing a comparison operation between the first meaning score and the second meaning score; and

if the first meaning score is higher than the second meaning score, then mapping the first text-based word to the first concept, or

if the first meaning score is not higher than the second meaning score, then mapping the first text-based word to the first concept.

5. The computer-implemented method of claim 4, wherein compositing the first meaning score comprises:

calculating a geometric mean score, a word-to-word score, and a sense-to-sense score; and

performing a weighted summation operation on the geometric score, the word-to-word score, and the sense-to-sense score.

6. The computer-implemented method of claim 1, wherein the plurality of topics include a first topic and a second topic, the first topic include a first concept, the second topic includes a second concept, and calculating the topic relevance rating comprises computing a function that is based on at least one of matches between the first concept and the second concept, the position of the concept 1 in the topic 1, and the position of the concept 2 in the topic 2.

7. The computer-implemented method of claim 1, further comprising determining that a first topic relevance rating for a first topic included in the plurality of topics is less than a predetermined threshold, and, in response removing the first topic from the plurality of topics prior to ranking the plurality of topics.

8. The computer-implemented method of claim 1, further comprising performing one or more comparison operations between the plurality of topics and another plurality of topics that is associated with a search term to establish a relevance between the search term and the text segment.

9. The computer-implemented method of claim 1, further comprising performing one or more syntactic mapping operations between the plurality of topics and the text segment based on the syntactic information to synthesize a textural summary of the text segment.

10. A non-transitory computer-readable storage medium including instructions that, when executed by a processing unit, cause the processing unit to interpret text segments based on word sense by performing the steps of:

parsing a text segment to generate one or more text-based words and related syntactic information;

mapping, via a processor, each of the one or more text-based words to at least one concept included in a database and based on a semantic network that includes the at least one concept and one or more relevance ratings associated with the at least one concept, wherein each concept included in the semantic network is associated with a meaning and at least one word;

generating a plurality of topics based on the mappings and the syntactic information, wherein each topic includes one or more of the concepts included in the semantic network;

for each topic included in the plurality of topics, calculating a topic relevance rating between the topic and at least another topic included in the plurality of topics based on the relevance ratings between the one or more concepts included in the topic and one or more concepts included in the another topic;

ranking the plurality of topics based on the topic relevance ratings to generate a ranked topic list; and

outputting an element for display based on the ranked topic list.

11. The non-transitory computer-readable storage medium of claim 10, wherein the semantic network includes a first concept and a second concept, the first concept is associated with a first word that is defined in a dictionary, and the second concept is associated with a second word that is a proper noun and is not defined in the dictionary.

12. The non-transitory computer-readable storage medium of claim 10, wherein the semantic network includes a first concept, the first concept is associated with a first phrase, and the first phrase includes a first word and a second word.

13. The non-transitory computer-readable storage medium of claim 10, wherein mapping each of the one or more text-based words comprises performing one or more word sense disambiguation operations.

14. The non-transitory computer-readable storage medium of claim 13, wherein the one or more word sense disambiguation operations for a first text-based word included in the one or more text-based words comprises:

compositing a first meaning score for the first text-based word based on a first concept included in the at least one concept;

compositing a second meaning score for the first text-based word based on a second concept included in the at least one concept;

performing a comparison operation between the first meaning score and the second meaning score; and

if the first meaning score is higher than the second meaning score, then mapping the first text-based word to the first concept, or

if the first meaning score is not higher than the second meaning score, then mapping the first text-based word to the second concept.

15. The non-transitory computer-readable storage medium of claim 14, wherein compositing the first meaning score comprises:

calculating a geometric mean score, a word-to-word score, and a sense-to-sense score; and

performing a weighted summation operation on the geometric score, the word-to-word score, and the sense-to-sense score.

16. The non-transitory computer-readable storage medium of claim 10, wherein the semantic network includes a first concept and the first concept is associated with both a first word and a second word that is a synonym for the first word.

17. The non-transitory computer-readable storage medium of claim 10, wherein calculating the topic relevance rating between the topic and the another topic included in the plurality of topics comprises:

ordering the concepts included in the topic and the another topic based on the related syntactic information; and

performing one or more comparison operations between the one or more concepts included in the topic and the one or more concepts included in the another topic based on at least one of syntactic types of the one or more concepts, semantic types of the one or more concepts, positions of the one or more concepts, the size of the topic and the size of another topic, and the relative difference between the topic and the another topic.

18. The non-transitory computer-readable storage medium of claim 10, wherein calculating the topic relevance rating between the topic and the another topic included in the plurality of topics comprises performing one or more comparison operations between a subset of the one or more concepts included in the topic and a subset of the one or more concepts included in the another topic.

19. A system configured to interpret text segments based on word sense, the system comprising:

a first memory that includes a semantic network, wherein the semantic network includes at least one concept included in a database and one or more relevance ratings associated with the at least one concept, and each concept included in the semantic network is associated with a meaning and at least one word; and

a processing unit configured to;

parse a text segment to generate one or more text-based words and related syntactic information;

map each of the one or more text-based words to the at least one concept included in the semantic network;

generate a plurality of topics based on the mappings and the syntactic information, wherein each topic includes one or more of the concepts included in the semantic network;

for each topic included in the plurality of topics, calculating a topic relevance rating between the topic and at least another topic included in the plurality of topics based on the relevance ratings between the one or more concepts included in the topic and one or more concepts included in the another topic;

ranking the plurality of topics based on the topic relevance ratings to generate a ranked topic list; and

outputting an element for display based on the ranked topic list.

20. The system of claim 19, wherein the semantic network includes a first concept and a second concept, the first concept is associated with a first word that is defined in a dictionary, and the second concept is associated with a second word that is a proper noun and is not defined in the dictionary.