Neural network for search retrieval and ranking

Abstract

Described herein is a mechanism for utilizing a neural network to identify and rank search results. A machine learning model is trained by converting training data comprising query-document entries into query term-document entries. The query term-document entries are utilized to train the machine learning model. A set of query terms are identified. The query terms can be derived from a query history. The trained machine learning model is used to calculate document ranking scores for the query terms and the resultant scores are stored in a pre-calculated term-document index. A query to search the document index is broken down into its constituent terms and an aggregate document ranking score is calculated from a weighted sum of the document ranking scores corresponding to the individual query terms. Because the term-document index can be pre-calculated, it can be downloaded to provide deep learning search capabilities in a computationally limited environment.

What is claimed is: 
     
         1 . A computing system that outputs a ranked list of search results in response to receipt of a query, the computing system comprising:
 a processor; and   memory storing instructions that, when executed by the processor, cause the processor to perform acts comprising:
 receiving a query, where the query comprises a first term and a second term; 
 identifying a document based upon the query; 
 subsequent to identifying the document, searching an index based upon the document and the first term; 
 determining that the index fails to include an entry for the document and the first term; 
 in response to determining that the index fails to include the entry for the document and the first term, providing the document and the first term to a computer-implemented model, where the computer-implemented model outputs a first ranking score based upon the document and the first term; 
 computing an overall ranking score for the document with respect to the query, wherein the overall ranking score is based upon an aggregate of the first ranking score and a second ranking score, where the second ranking score is assigned to the document and the second term in the query; and 
 returning the ranked list of documents based upon the query, where the document is positioned in the ranked list of documents based upon the overall ranking score. 
   
     
     
         2 . The computing system of  claim 1 , the acts further comprising:
 subsequent to identifying the document, retrieving the second ranking score from the index, where the second ranking score corresponds to the document and the second term in the index, and further where the second ranking score was previously output by the computer-implemented model based upon the document and the second term.   
     
     
         3 . The computing system of  claim 1 , the acts further comprising:
 subsequent to identifying the document, searching the index based upon the document and the second term;   determining that the index fails to include an entry for the document and the second term;   in response to determining that the index fails to include the entry for the document and the second term, providing the document and the second term to the computer-implemented model, where the computer-implemented model outputs the second ranking score based upon the document and the second term.   
     
     
         4 . The computing system of  claim 3 , the acts further comprising updating the index to include the second ranking score, where the second ranking score is assigned to the document and the second term in the index. 
     
     
         5 . The computing system of  claim 1 , the acts further comprising updating the index to include the first ranking score, where the first ranking score is assigned to the document and the first term in the index. 
     
     
         6 . The computing system of  claim 1 , wherein the computer-implemented model is a neural network. 
     
     
         7 . The computing system of  claim 1 , where the query includes a third term, the method further comprising:
 subsequent to identifying the document, searching an index based upon the document and the third term;
 determining that the index fails to include an entry for the document and the third term; and 
 in response to determining that the index fails to include the entry for the document and the third term, providing the document and the third term to the computer-implemented model, where the computer-implemented model outputs a third ranking score based upon the document and the first term, and further where the overall ranking score is based upon an aggregate of the first ranking score, the second ranking score, and the third ranking score. 
   
     
     
         8 . The computing system of  claim 1 , wherein the index comprises a plurality of ranking scores for the document, where each ranking score for the document corresponds to a different term. 
     
     
         9 . A method for outputting a ranked list of documents based upon a query, the method performed by a computing system, the method comprising:
 receiving a query, where the query comprises a first term and a second term;   identifying a document based upon the query;   obtaining a first ranking score, where the first ranking score is generated by a computer-implemented model based upon the document and the first term of the query;   obtaining a second ranking score, where the second ranking score is generated by the computer-implemented model based upon the document and the second term of the query;   computing an overall ranking score for the document with respect to the query, where the overall ranking score is based upon an aggregate of the first ranking score and the second ranking score; and   outputting the ranked list of documents based upon the query, where the document is positioned in the ranked list of documents based upon the overall ranking score for the document.   
     
     
         10 . The method of  claim 9 , wherein obtaining the first ranking score comprises retrieving the first ranking score from an index, where the first ranking score is mapped to a combination of the document and the first term in the index. 
     
     
         11 . The method of  claim 10 , wherein obtaining the second ranking score comprises retrieving the second ranking score from the index, where the second ranking score is mapped to a combination of the document and the second term in the index. 
     
     
         12 . The method of  claim 10 , wherein obtaining the second ranking score comprises providing the document and the second term to the computer-implemented model, where the computer-implemented model outputs the second ranking score based upon the document and the second term. 
     
     
         13 . The method of  claim 9 , wherein obtaining the first ranking score comprises providing the document and the first term to the computer-implemented model, where the computer-implemented model outputs the first ranking score based upon the document and the first term. 
     
     
         14 . The method of  claim 13 , wherein obtaining the second ranking score comprises providing the document and the second term to the computer-implemented model, where the computer-implemented model outputs the second ranking score based upon the document and the second term. 
     
     
         15 . The method of  claim 9 , wherein the computer-implemented model is a neural network. 
     
     
         16 . Machine storage media comprising instructions that, when executed by a processor, cause the processor to perform acts comprising:
 receiving a query, where the query comprises a first term and a second term;   identifying a document based upon the query;   subsequent to identifying the document, searching an index based upon the document and the first term;   determining that the index fails to include an entry for the document and the first term;   in response to determining that the index fails to include the entry for the document and the first term, providing the document and the first term to a computer-implemented model, where the computer-implemented model outputs a first ranking score based upon the document and the first term;   computing an overall ranking score for the document with respect to the query, wherein the overall ranking score is based upon an aggregate of the first ranking score and a second ranking score, where the second ranking score is assigned to the document and the second term in the query; and   returning a ranked list of documents based upon the query, where the document is positioned in the ranked list of documents based upon the overall ranking score.   
     
     
         17 . The machine storage medium of  claim 16 , the acts further comprising:
 subsequent to identifying the document, retrieving the second ranking score from the index, where the second ranking score corresponds to the document and the second term in the index, and further where the second ranking score was previously output by the computer-implemented model based upon the document and the second term.   
     
     
         18 . The machine storage medium of  claim 16 , the acts further comprising:
 subsequent to identifying the document, searching the index based upon the document and the second term;   determining that the index fails to include an entry for the document and the second term;   in response to determining that the index fails to include the entry for the document and the second term, providing the document and the second term to the computer-implemented model, where the computer-implemented model outputs the second ranking score based upon the document and the second term.   
     
     
         19 . The machine storage medium of  claim 18 , the acts further comprising updating the index to include the second ranking score, where the second ranking score is assigned to the document and the second term in the index. 
     
     
         20 . The machine storage medium of  claim 16 , the acts further comprising updating the index to include the first ranking score, where the first ranking score is assigned to the document and the first term in the index.