Document expansion in speech retrieval
View Patent ↗Methods of document expansion for a speech retrieval document by a recognizer. A database of vectors of automatic transcriptions of documents is accessed and the vectors are truncated by removing all terms that are not recognizable by the recognizer to create truncated vectors. Terms in the vectors are then weighted to associate the truncated vectors with the untruncated vectors. Terms not recognized by the recognizer are then added back to the weighted, truncated vectors. The retrieval effectiveness may then be measured.
1. A method of evaluating effectiveness of speech retrieval of documents, the method comprising:
removing terms in vectors that are not recognized by a recognizer, the vectors being associated with automatic transcriptions of documents;
generating weighted vectors by modifying weights of terms in the vectors;
adding to the weighted vectors terms which were not recognized by the recognizer;
receiving a plurality of speech queries for the documents, wherein the speech queries are based on the weighted vectors; and
determining effectiveness of the received plurality of speech queries based on queries in the plurality of speech queries which result in a number of relevant documents equal to or greater than a predetermined threshold.
2. The method of claim 1 , wherein the documents relate to a first document based at least on a frame of time.
3. The method of claim 2 , wherein the documents relate to the first document based at least on the fact that the first document and the documents both relate to a news corpus.
4. The method of claim 2 , the method further comprising modifying the first document by expanding the document to be larger than its original length.
5. The method of claim 2 , the method further comprising adding at least one word from the documents to the first document.
6. The method of claim 5 , wherein adding the at least one word further comprises identifying frequently occurring words from the documents and adding the frequently occurring words to the first document.
7. The method of claim 2 , the method further comprising running an automatic transcription of the first document as a query on the collection of documents to retrieve related documents from the documents.
8. The method of claim 7 , wherein the automatic transcription of the first document is weighted when used as the query.
9. A non-transitory computer-readable storage medium storing instructions which, when executed by a computing device, cause the computing device to evaluate effectiveness of speech retrieval of documents, the instructions comprising:
removing terms in vectors that are not recognized by a recognizer, the vectors being associated with automatic transcriptions of documents;
generating weighted vectors by modifying weights of terms in the vectors;
adding to the weighted vectors terms which were not recognized by the recognizer;
receiving a plurality of speech queries for the documents, wherein the speech queries are based on the weighted vectors; and
determining effectiveness of the received plurality of speech queries based on queries in the plurality of speech queries which result in a number of relevant documents equal to or greater than a predetermined threshold.
10. The non-transitory computer-readable storage medium of claim 9 , wherein documents relates to a first document based at least on a frame of time.
11. The non-transitory computer-readable storage medium of claim 10 , wherein the documents relates to the first document based at least on the fact that the first document and the documents both relate to a news corpus.
12. The non-transitory computer-readable storage medium of claim 10 , the method further comprising adding at least one word from the documents to the first document.
13. The non-transitory computer-readable storage medium of claim 12 , wherein adding the at least one word further comprises identifying frequently occurring words from the documents and adding the frequently occurring words to the first document.
14. The non-transitory computer-readable storage medium of claim 10 , the instructions further comprising running an automatic transcription of the first document as a query on the collection of documents to retrieve related documents from the documents.
15. The non-transitory computer-readable storage medium of claim 14 , wherein the automatic transcription of the first document is weighted when used as the query.
16. A system for evaluating effectiveness of speech retrieval of documents, the system comprising:
removing terms in vectors that are not recognized by a recognizer, the vectors being associated with automatic transcriptions of documents;
generating weighted vectors by modifying weights of terms in the vectors;
adding to the weighted vectors terms which were not recognized by the recognizer;
receiving a plurality of speech queries for the documents, wherein the speech queries are based on the weighted vectors; and
determining effectiveness of the received plurality of speech queries based on queries in the plurality of speech queries which result in a number of relevant documents equal to or greater than a predetermined threshold.