# US 2013/0159320 A1 · Clickthrough-based latent semantic model > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2013/0159320 A1 - Document ID: US20130159320A1 - Status: Published application - Kind: A1 - Application number: 13/329,345 - Filed: 2011-12-19 - Published: 2013-06-20 - Earliest claimed: 2011-12-19 - Assignees: Microsoft Corporation - Inventors: Jianfeng Gao; Kristina Toutanova; Wen-tau Yih - Web page: https://patentgrep.com/patent/US20130159320A1 ## CPC - G06F 16/9535: Search customisation based on user profiles and personalisation - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract There is provided a computer-implemented method and system for ranking documents. The method includes identifying a number of query-document pairs based on clickthrough data for a number of documents. The method also includes building a latent semantic model based on the query-document pairs and ranking the documents for a search based on the latent semantic model. ## Claims ### Claim 1 A computer-implemented method for ranking documents, comprising: identifying a plurality of query-document pairs based on clickthrough data for a plurality of documents; building a latent semantic model based on the plurality of query-document pairs, wherein the plurality of query-document pairs comprises a plurality of query-title pairs, and wherein building the latent semantic model comprises building a bilingual topic model by using the plurality of query-title pairs to learn a semantic representation of a query based on a likelihood that the query is a semantics-based translation of each of the plurality of documents; and ranking the plurality of documents for a Web search based on the latent semantic model. ### Claim 2 (depends on 1) The method of claim 1, wherein ranking the plurality of documents based on the latent semantic model comprises reordering the plurality of documents for the search based on a relevance of each of the plurality of documents to a query. ### Claim at position 3 3-4. (canceled) ### Claim 5 (depends on 3) The method of claim 3, comprising ranking the plurality of documents for the search based on the likelihood that the query is the semantics-based translation of each of the plurality of documents. ### Claim 6 (depends on 1) The method of claim 1, wherein building the latent semantic model based on the plurality of query-document pairs comprises building a discriminative projection model by using the plurality of query-document pairs to learn a projection matrix. ### Claim 7 (depends on 6) The method of claim 6, comprising generating vector representations of a query and a title of each of the plurality of documents, wherein the vector representations comprise sparse, high-dimensional term vectors. ### Claim 8 (depends on 7) The method of claim 7, comprising ranking the plurality of documents for the search based on a distance between the vector representations of the query and the title of each of the plurality of documents within the semantic space, wherein the projection matrix is used to map the vector representations of the query and the title of each of the plurality of documents to a semantic space, wherein the semantic space comprises a dense, low-dimensional space. ### Claim 9 (depends on 1) The method of claim 1, wherein a query and a paired title of one of the plurality of documents comprise a common distribution of topics, and wherein the query and the paired title use different vocabularies or language styles, or both, to express the common distribution of topics. ### Claim 10 A system for ranking documents, comprising: a processor that is adapted to execute stored instructions; and a system memory, wherein the system memory comprises code configured to: identify a plurality of query-title pairs based on clickthrough data for a plurality of documents; build a latent semantic model based on the plurality of query-title pairs by building a bilingual topic model by learning a semantic representation of a query based on a likelihood that the query is a semantics-based translation of each of the plurality of documents; and rank the plurality of documents for a search based on the latent semantic model. ### Claim 11 (canceled) ### Claim 12 (depends on 10) The system of claim 10, wherein the system memory comprises code configured to identify the plurality of query-title pairs using posterior regularization. ### Claim 13 (depends on 10) The system of claim 10, wherein the system memory comprises code configured to identify the hidden topics or semantic representation of the plurality of query-title pairs in a language-independent manner using maximum a posteriori (MAPS estimation techniques. ### Claim 14 (depends on 10) The system of claim 10, wherein the system memory comprises code configured to rank the plurality of documents for the search based on the likelihood that the query is the semantics-based translation of each of the plurality of documents. ### Claim 15 (depends on 10) The system of claim 10, wherein the query and a paired title of one of the plurality of documents comprise a common distribution of topics, and wherein the query and the paired title use different vocabularies or language styles, or both, to express the common distribution of topics. ### Claim 16 One or more non-transitory, computer-readable storage media for storing computer-readable instructions, the computer-readable instructions providing a document ranking system when executed by one or more processing devices, the computer-readable instructions comprising code configured to: identify a plurality of query-title pairs based on clickthrough data for a plurality of documents; build a latent semantic model by using the plurality of query-title pairs to learn a projection matrix, wherein the latent semantic model is built by building a bilingual topic model to learn a semantic representation of a query based on a likelihood that the query is a semantics-based translation of each of the plurality of documents; map vector representations of a query and a title of each of the plurality of documents to a semantic space using the projection matrix; and rank the plurality of documents for a search based on a distance between the vector representations of the query and the title of each of the plurality of documents within the semantic space. ### Claim 17 (depends on 16) The one or more non-transitory, computer-readable storage media of claim 16, wherein the latent semantic model comprises a discriminative projection model. ### Claim 18 (depends on 16) The one or more non-transitory, computer-readable storage media of claim 16, wherein the vector representations comprise sparse, high-dimensional term vectors, and wherein the semantic space comprises a dense, low-dimensional space. ### Claim 19 (depends on 16) The one or more non-transitory, computer-readable storage media of claim 16, wherein the computer-readable instructions comprise code configured to use the projection matrix to map the vector representations to the semantic space through a matrix multiplication procedure. ### Claim 20 (depends on 16) The one or more non-transitory, computer-readable storage media of claim 16, wherein the distance between the vector representations within the semantic space comprises a measure of a relevance of each of the plurality of documents to the query.