Back to results
US 9,436,674 B2Granted
Signal processing approach to sentiment analysis for entities in documents
AssigneeAttivio, Inc.
InventorJohn O'Neil
Life of this patent
Estimating the term…
Abstract
A document can be processed to provide sentiment values for phrases in the document. The sequence of sentiment values associated with the sequence of phrases in a document can be handled as if they were a sampled discrete time signal. For phrases which have been identified as entities, a filtering operation can be applied to the sequence of sentiment values around each entity to determine a sentiment value for the entity.
Claims20
A computer system comprising:
- a non-transitory computer readable storage in which data are stored, the data including:
- data identifying entities described in a document, stored in computer memory, wherein the document includes a plurality of tokens appearing in an order in the document, and
- data defining a sequence of sentiment values assigned to the tokens in the document, the sequence of sentiment values being in the order the corresponding tokens appear in the document;
- and
- at least one processor for processing the data in the computer readable storage to assign a sentiment value to one of the identified entities in the document by:
- applying a filter to the sequence of sentiment values in the non-transitory computer readable storage, the filter having a width defined by a number of tokens, such that the sentiment value of a token is spread among other nearby tokens within the width of the filter, whereby each entity has contributions due to sentiment values from nearby tokens within the width of the filter, and
- for each entity in the document, combining the contributions from the nearby tokens to produce a sentiment value for the entity.
4 more elements
A computer program product, comprising:
- a non-transitory computer readable storage device storing computer program instructions which, when processed by a computer system including at least one processor and a computer memory, enable the at least one processor to:
- access, in the computer memory, data identifying entities described in a document, wherein the document includes a plurality of tokens appearing in an order in the document, and data defining a sequence of sentiment values assigned to the tokens in the document, the sequence of sentiment values being in the order the corresponding tokens appear in the document,
- apply a filter to the sequence of sentiment values in the computer memory, the filter having a width defined by a number of tokens, such that the sentiment value of a token is spread among other nearby tokens within the width of the filter, whereby each entity has contributions due to sentiment values from nearby tokens within the width of the filter, and
- combine, for each entity in the document, the contributions from the nearby tokens to produce a sentiment value for the entity.
1 more element
A computer program product, comprising:
- a non-transitory computer readable storage device storing computer program instructions which, when processed by a computer system including at least one processor and a computer memory, enable the at least one processor to:
- access the non-transitory computer readable storage in which data are stored, the data including data identifying entities described in a document, wherein the document includes a plurality of tokens appearing in an order in the document, and data defining sentiment values assigned to the tokens in the document, and
- process, using the at least one processor, the data in the non-transitory computer readable storage to assign a sentiment value to one of the identified entities in the document by applying a filter to a sequence of the sentiment values corresponding to a sequence of the tokens in the order the tokens appear in the document, the filter having a width defined by a number of tokens, the entity having a position in the sequence of tokens, the filter providing a combination of contributions of the sentiment values associated with the tokens surrounding the position of the entity in the document within the width of the filter.
Dependencies are inferred from the claim text.