System and method for customized sentiment signal generation through machine learning based streaming text analytics

• US 9,514,133 B1
• Filed: 08/21/2013
• Issued: 12/06/2016
• Est. Priority Date: 06/25/2013
• Status: Expired

• Current Assignee: JP Morgan Chase Bank, N.A. (JP Morgan Chase & Co)
• Original Assignee: JP Morgan Chase Bank, N.A. (JP Morgan Chase & Co)
• Inventors: Kursun, Eren
• Primary Examiner(s): Kuddus, Daniel
• Application Number:
  US13/972,388
• Time in Patent Office: 1,203 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 16/24568 Data stream processing; Continuous queries
• CPC Class Codes: G06F 16/285 Clustering or classification
• CPC Class Codes: G06F 16/367 Ontology
• CPC Class Codes: G06F 16/9535 Search customisation based on user profiles and personalisation
• CPC Class Codes: G06F 40/134 Hyperlinking
• CPC Class Codes: G06F 40/279 Recognition of textual entities
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06F 40/40 Processing or translation of natural language natural language analysis G06F40/20; semantic analysis G06F40/30
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models

Abstract

Systems and methods may provide customized integrated indexes and visualization. Sentiment analytics may be based on natural language processing techniques. Users may select from among a range of indexes that reflect a variety of sources. Text scoring metrics or indices may incorporate frequency of mention, link to broker action, sentence location of first mention, etc. Depending on the temporal and sentiment characteristics of interest, the user may select from a range of news sources, research reports, analysts, social media sources, and may assign a customized weight value to each source. The scores may then be merged. After scoring, the user may be presented with news links directly from sentiment indexes (e.g., from top ranking in terms of sentiment scores, etc.). Advanced visualization capabilities may provide output for users to assist in decision making processes.

Claims

1. A computer implemented method, comprising:

extracting, by the computer processor, one or more sentiment signals comprising weighted, customized sentiment metrics for words and expressions from a library comprising one or more research reports from one or more sources;

calculating normalized profiles for the one or more sources of data based on one or more of geography, sector, analyst, company, and streaming real time data feedback;

normalizing, by the computer processor, the one or more sentiment signals based on text positioning, sentence structure, and data source;

calculating an overall sentiment score;

receiving, by the computer processor, one or more data streams;

receiving, by the computer processor, a set of weighted, customized metrics based on the one or more sentiment signals wherein the set of weighted, customized metrics is received from a dynamically updated database that assigns and updates sentiments based on empirical data, customizes factors based on analyst, sectors, and geography, and comprises numeric sensitivity factors for numeric expressions;

applying the set of weighted, customized metrics to the one or more data streams;

outputting, by the computer processor, a customized data stream that is a result of the application of the set of weighted, customized metrics;

ranking the one or more sentiment signals based on the overall sentiment score; and

providing a visualization of the ranking including a link to each source associated with the one or more sentiment signals.

2. The method of claim 1, further comprising:

extracting, by the computer processor, one or more second sentiment signals comprising one or more of numeric expressions, acronyms, and symbols; and

linking, by the computer processor, the one or more second sentiment signals to words and expressions to which sentiment of the one or more second sentiment signals refers to.

3. The method of claim 1, wherein the research reports are related to a financial institution.

4. The method of claim 1, wherein the one or more sentiment signals are extracted temporally based on a structure and context of a sentence containing the sentiment signal.

5. The method of claim 1, wherein each word or expression has one or more sentiment metrics that have a sentiment that varies temporally.

6. The method of claim 1, wherein each word or expression has one or more sentiment metrics that have a sentiment that varies contextually.

7. The method of claim 1, wherein the one or more sentiment signals span two or more sources of data and are merged based on entity characteristics and a look-up table.

8. The method of claim 1, wherein the normalized profiles comprise a vector of factors.

9. The method of claim 1, wherein the one or more sentiment signals are dynamically weighted and polarized based on empirical data.

10. The method of claim 1, wherein the one or more sentiment signals pertain to more than one entity.

11. The method of claim 10, further comprising:

combining each of the one or more sentiment signals for the more than one entity.

12. The method of claim 1, wherein the one or more sentiment signals span across two or more sources of data in the library.

13. The method of claim 12, wherein the one or more sentiment signals pertain to more than one entity.

14. The method of claim 13, further comprising:

combining each of the one or more sentiment signals for the more than one entity.

15. The method of claim 12, further comprising:

performing a cross check of the one or more sentiment signals for consistency within the library.

16. The method of claim 1, wherein the one or more sentiment signals are extracted across a hierarchy within a given taxonomy of entities.

17. The method of claim 1, wherein the one or more data streams comprise reports, streaming news, and social media.

18. The method of claim 1, wherein the numeric sensitivity factor is dynamically updated on a per entity basis.

19. The method of claim 1, wherein the outputting comprises multi-dimensional and temporal visualization of the sentiment.

20. The method of claim 1, wherein the outputting comprises a visualization of the customized data stream.

21. The method of claim 20, wherein the visualization comprises a composite signal based on the one or more data streams.

22. The method of claim 20, wherein the visualization comprises rendering the customized data stream separately for each of the one or more data streams, wherein each separate rendered data stream is color coded.

23. A computer implemented method, comprising:

extracting, by computer processor, one or more sentiment signals comprising a set of weighted, customized sentiment metrics for words and expressions from a source document;

calculating a normalized profile for the source document based on one or more of geography, sector, analyst, and company;

cross-checking each of the one or more sentiment signals for consistency within the source document;

applying the normalized profile to the one or more sentiment signals;

normalizing, by the computer processor, the one or more sentiment signals based on text positioning, sentence structure, and document type;

calculating an overall sentiment score;

receiving, by the computer processor, one or more data streams;

receiving, by the computer processor, a set of weighted, customized metrics based on the one or more sentiment signals wherein the set of weighted, customized metrics is received from a dynamically updated database that assigns and updates sentiments based on empirical data, customizes factors based on analyst, sectors, and geography, and comprises numeric sensitivity factors for numeric expressions;

applying the set of weighted, customized metrics to the one or more data streams;

outputting, by the computer processor, a customized data stream that is a result of the application of the set of weighted, customized metrics; and

ranking the one or more sentiment signals based on the overall sentiment score; and

providing a visualization of the ranking including a link to each source associated with the one or more sentiment signals.

24. A computer implemented method, comprising:

extracting, by the computer processor, one or more sentiment signals comprising weighted, customized sentiment metrics for words and expressions from a live streaming data source comprising one or more sources of data delivered over a computer network;

calculating normalized profiles for the one or more sources of data based on one or more of geography, sector, analyst, company, and streaming real time data feedback;

normalizing, by the computer processor, the one or more sentiment signals based on text positioning, sentence structure, and data source;

calculating an overall sentiment score;

receiving, by the computer processor, one or more data streams;

receiving, by the computer processor, a set of weighted, customized metrics based on the one or more sentiment signals wherein the set of weighted, customized metrics is received from a dynamically updated database that assigns and updates sentiments based on empirical data, customizes factors based on analyst, sectors, and geography, and comprises numeric sensitivity factors for numeric expressions;

applying the set of weighted, customized metrics to the one or more data streams;

outputting, by the computer processor, a customized data stream that is a result of the application of the set of weighted, customized metrics;

ranking the one or more sentiment signals based on the overall sentiment score; and

providing a visualization of the ranking including a link to each source associated with the one or more sentiment signals.

25. The method of claim 24, further comprising:

extracting, by the computer processor, one or more second sentiment signals comprising one or more of numeric expressions, acronyms, and symbols; and

linking, by the computer processor, the one or more second sentiment signals to words and expressions to which sentiment of the one or more second sentiment signals refers to.

26. The method of claim 24, wherein the one or more sentiment signals are extracted temporally based on a structure and context of a sentence containing the sentiment signal.

27. The method of claim 24, wherein each word or expression has one or more sentiment metrics that have a sentiment that varies temporally.

28. The method of claim 24, wherein each word or expression has one or more sentiment metrics that have a sentiment that varies contextually.

29. The method of claim 24, wherein the one or more sentiment signals span two or more sources of data and are merged based on entity characteristics and a look-up table.

30. The method of claim 24, wherein the normalized profiles comprise a vector of factors.

31. The method of claim 24, wherein the one or more sentiment signals are dynamically weighted and polarized based on empirical data.

32. The method of claim 24, wherein the one or more sentiment signals pertain to more than one entity.

33. The method of claim 32, further comprising:

combining each of the one or more sentiment signals for the more than one entity.

34. The method of claim 24, wherein the one or more sentiment signals span across two or more sources of data in the streaming data source.

35. The method of claim 34, wherein the one or more sentiment signals pertain to more than one entity.

36. The method of claim 35, further comprising:

combining each of the one or more sentiment signals for the more than one entity.

37. The method of claim 34, further comprising:

performing a cross check of the one or more sentiment signals for consistency within the streaming data source.

38. The method of claim 24, wherein the one or more sentiment signals are extracted across a hierarchy within a given taxonomy of entities.

39. The method of claim 24, wherein the one or more data streams comprise reports, streaming news, and social media.

40. The method of claim 24, wherein the numeric sensitivity factor is dynamically updated on a per entity basis.

41. The method of claim 24, wherein the outputting comprises multi-dimensional and temporal visualization of the sentiment.

42. The method of claim 24, wherein the outputting comprises a visualization of the customized data stream.

43. The method of claim 42, wherein the visualization comprises rendering the customized data stream separately for each of the one or more data streams, wherein each separate rendered data stream is color coded.

44. The method of claim 24, wherein the visualization comprises a composite signal based on the one or more data streams.

45. A system, comprising:

a processor; and

a memory comprising computer-readable instructions which when executed by the processor cause the processor to perform the steps comprising;

extracting one or more sentiment signals comprising weighted, customized sentiment metrics for words and expressions from a library comprising one or more sources of data;

calculating normalized profiles for the one or more sources of data based on one or more of geography, sector, analyst, company, and streaming real time data feedback;

normalizing the one or more sentiment signals based on text positioning, sentence structure, and document type;

calculating an overall sentiment score;

receiving one or more data streams;

receiving a set of weighted, customized metrics based on the one or more sentiment signals wherein the set of weighted, customized metrics is received from a dynamically updated database that assigns and updates sentiments;

applying the set of weighted, customized metrics to the one or more data streams;

outputting a customized data stream that is a result of the application of the set of weighted, customized metrics;

ranking the one or more sentiment signals based on the overall sentiment score; and

providing a visualization of the ranking including a link to each actual source associated with the one or more sentiment signals.

46. The system of claim 45, further comprising:

extracting one or more second sentiment signals comprising one or more of numeric expressions, acronyms, and symbols; and

linking the one or more second sentiment signals to words and expressions to which sentiment of the one or more second sentiment signals refers to.

47. The system of claim 45, wherein the normalized profiles comprise a vector of factors.

49. The system of claim 45, wherein the one or more data streams comprise reports, streaming news, and social media.

50. The system of claim 45, wherein the sources of data comprise research reports.

51. The system of claim 50, wherein the research reports are related to a financial institution.

52. The system of claim 45, wherein the one or more sentiment signals is dynamically weighted and polarized based on empirical data, wherein further the weights and polarity are dynamically updated.

53. The system of claim 45, wherein the one or more sentiment signals pertains to more than one entity.

54. The system of claim 45, wherein the one or more sentiment signals span across two or more research reports in the library.

56. The system of claim 45, wherein the dynamically updated database assigns and updates sentiments based on empirical data comprising market data.

57. The system of claim 45, wherein the dynamically updated database customizes factors based on analyst, sectors, and geography.

58. The system of claim 45, wherein the dynamically updated database comprises numeric sensitivity factors for different keywords and expressions.

59. The system of claim 45, wherein the dynamically updated database captures customized keywords and expressions related to financial and investment applications.

62. The system of claim 45, wherein the terms are stored in a domain specific manner.

63. The system of claim 45, wherein the dynamically weighted dictionary associated weights with its contents based on dynamic factors that are internal and external to the system.

64. The system of claim 45, wherein the sentiment is a dynamically calculable or queriable weighted taxomony and the source attributes are descriptive and are configured to allow a graphical traversal of all attributes, sentiment, and concepts.

65. The system of claim 45, wherein the object has a lifespan comprising a set period of time and the object is after the set period of time.

48. The method of 45, wherein the weighted, customized metrics are updated in real time in the dynamically updated database through machine learning.

55. The system of 45, further comprising:

performing a cross check of the one or more sentiment signals within the library for consistency.

60. A system, comprising:

a processor; and

a memory comprising computer-readable instructions which when executed by the processor cause the processor to perform the steps comprising;

parsing one or more sources of data for sentiment and source attributes;

extracting source attributes from the one or more sources of data;

fusing the source attributes with the sentiment;

linking the sentiment information with its original source, related content, and other related material;

weighting the sentiment information based on importance and relevance;

storing, in a dynamically weighted dictionary, terms associated with sentiment and concept;

outputting an object with dynamic sentiment and attributes;

receiving one or more data streams;

receiving a set of weighted, customized metrics based on one or more sentiment signals wherein the set of weighted, customized metrics is received from a dynamically updated database that assigns and updates sentiments based on empirical data, customizes factors based on analyst, sectors, and geography, and comprises numeric sensitivity factors for numeric expressions;

applying the set of weighted, customized metrics to the one or more data streams; and

outputting a customized data stream that is a result of the application of the set of weighted, customized metrics.

61. The system of claim 60, wherein the one or more sources of data comprise one or more of research reports, the Internet, and social media.