Browse Prior Art Database

State aware Text Analytics Disclosure Number: IPCOM000236364D
Publication Date: 2014-Apr-22
Document File: 3 page(s) / 120K

Publishing Venue

The Prior Art Database


A method for state aware text analytics is disclosed.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 3

State aware Text Analytics

Disclosed is a method for state aware text analytics.

The industry standard Unstructured Information Management Architecture (UIMA) provides a series of rich features for performing text analytics on unstructured information(text). The architecture proscribes techniques that can be used for individual words or phrases and relationships between words or phrases at the sentence, paragraph, and document level in unstructured information. However, creating natural language processing models that take the state of a situation into consideration is not possible in today's architecture. For example, if one were processing documents associated with traffic accidents involving chemical compounds that became unstable at particular temperatures, the severity of an accident could only be determined if the air temperature and immediate local forecast were known. Similarly, if a question answering system were being used to suggest a potential treatment in a healthcare situation, the "state" of the patient, what other treatments have been tried, other medical conditions, the patient's chronological medical history, etc might be key in suggesting a new treatment plan.

In another example, a client, in the media and entertainment, industry keeps detailed logs of sporting events. The text in the logs are entered by production assistants as they watch the games. Text describing each play is captured and each entry is time stamped. If the question answering system would be unable to create annotations for "clutch" plays based on the textual information from the game's log . Using the game's situation to make a decision on whether the play was "clutch" or not is not a documented client requirement. The exact same textual description of a play, a great catch in baseball or a highly contested made three pointer in basketball, might or might not be considered a clutch play depending on the situation of the game when the play occurred.

If the three pointer was made in early in the first half of a close game, most observers would conclude that while the play might be considered a great shot, it wouldn't be considered "clutch". However, the exact same three point shot, made in the last 30 seconds of a two point game would clearly be a "clutch" play. As such, in order to be able to correctly annotate the latter 3 point play's text as "clutch", the state of the game must be included in the annotation creation process.

This deficiency may be overcome by keeping attributes which define the state of the game inside a Common Analysis Structure (CAS) alongside individu...