# US 8,788,342 B2 · Intelligent feature expansion of online text ads > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 8,788,342 B2 - Document ID: US8788342B2 - Status: Granted - Kind: B2 - Application number: 12/935,283 - Filed: 2010-03-04 - Published: 2012-12-13 - Granted: 2014-07-22 - Earliest claimed: 2010-03-04 - Assignees: Yahoo! Inc. - Applicants: Jian Hu; Hao Zheng - Inventors: Jian Hu; Hao Zheng - Web page: https://patentgrep.com/patent/US8788342B2 ## CPC - G06Q 30/0244: {Optimization} - G06F 16/313: {Selection or weighting of terms for indexing} - G06F 16/3338: {Query expansion} - G06F 16/9535: Search customisation based on user profiles and personalisation - G06Q 30/0256: {User search} - G06Q 30/0276: {Advertisement creation} - G06Q: INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract Embodiments employ knowledge from network searches to enrich features useable to describe an advertisement. Advertisement content may be segmented to generate key phrases that are rank ordered based on a phrase's pointwise mutual information value with Ad keywords and term frequency-inverse document frequency (TF-IDF) scores. These key phrases are employed to generate a plurality of query searches that are performed to generate a plurality of search result snippets. The snippets are then analyzed and ranked to identify query expansion features having a TF-IDF value over a given threshold. A subset of these ranked phrases are combined with the Ad keywords to be used to rebuild an Ad index for the given advertisement. When content is to be delivered to a client device, content keywords are compared to the Ad index to select an advertisement to be provided with the content to the client device. ## Claims ### Claim 1 A network device configured to send and receive data over a network, the network device comprising: at least one non-transitory storage medium comprising at least one set of instructions for feature expansions of online advertisements; at least one processor in communication with the at least one non-transitory storage medium and configured to execute the at least one set of instructions to: receive an advertisement with Ad keywords associated with the advertisement; segment content of the advertisement into a plurality of phrases; perform a semantic analysis to the plurality of phrases; select top-ranked phrases from the plurality of phrases based on the semantic analysis to form a reduced subset of phrases generate a query list having a plurality of queries by combining each phrase in the reduced subset of phrases that has less than a defined number of words with another independent phrase in the reduced subset of phrases to create queries comprising two or more words; for each query within the query list, perform an internet web search to identify a set of descriptions, each description describing a search result from the internet web search; select a second set of phrases denoted as a query expansion feature set for each query based on an analysis of the set of descriptions; rank the second set of phrases; generate an expanded Ad index, wherein to generate the expanded Ad index the at least one processor is configured to: select from the ranked second set of phrases a subset of phrases as an ad expanded feature set; and combine the ad expanded feature set with the Ad keywords associated with the advertisement; and when in receipt of a request to provide to a client device a second content for display, select an advertisement from a plurality of advertisements based on the expanded Ad index. ### Claim 2 (depends on 1) The network device of claim 1, wherein to conduct the semantic analysis, the at least one processor is further configured to: group related of words from the content of the advertisement to generate a phrase from the content of the advertisement; determine a pointwise mutual information (PMI) value for each generated phrase with each Ad keyword; rank in order the determined PMI values based on a Term Frequency-Inverse Document Frequency (TF-IDF) score for each generated phrase; and select from the rank ordering a subset of phrases as the reduced subset of phrases for the advertisement. ### Claim 3 (depends on 1) The network device of claim 1, wherein the content of the advertisement comprises at least one of a title of the advertisement, a description of the advertisement, or a Uniform Resource Locator (URL). ### Claim 4 (depends on 1) The network device of claim 1, wherein to analyze the set of search result descriptions, the processor is further configured to identify a phrase with a Term Frequency-Inverse Document Frequency (TF-IDF) score for the second set of phrases above a threshold value, the second set of phrases being denoted as a query expansion feature set. ### Claim 5 (depends on 1) The network device of claim 1, wherein the second content for display further comprises at least one of a web page and a search result. ### Claim 6 A non-transitory computer-readable storage device having computer executable instructions stored thereon, the computer-executable instructions configured to cause a processor to perform acts of: receiving an advertisement with Ad keywords associated with the advertisement; performing a semantic analysis to the plurality of phrases; segmenting content of the advertisement into a plurality of phrases; selecting top-ranked phrases from the plurality of phrases based on the semantic analysis to form a reduced subset of phrases; generating a query list having a plurality of queries by combining each phrase in the reduced subset of phrases that has less than a defined number of words with another independent phrase in the reduced subset of phrases to create queries comprising two or more words; for each query within the query list, performing an internet web search to identify a set of descriptions, each description describing a search result from the internet web search; analyzing the set of descriptions to select a second set of phrases denoted as a query expansion feature set for each query; ranking the second set of phrases within each query expansion feature set and a frequency over all query expansion feature sets; generating an expanded Ad index by selecting from the ranked second set of phrases a subset of phrases as an ad expanded feature set that is combined with the Ad keywords associated with the advertisement; and when receiving a second content for display at a client device, selecting an advertisement from a plurality of advertisements based on the expanded Ad index. ### Claim 7 (depends on 6) The non-transitory computer-readable storage medium of claim 6, wherein conducting the semantic analysis comprises: grouping related words from the content of the advertisement to generate a phrase from the content of the advertisement; determining a pointwise mutual information (PMI) value for each generated phrase with each Ad keyword; rank ordering the determined PMI values based on a Term Frequency-Inverse Document Frequency (TF-IDF) score for each generated phrase; and selecting from the rank ordering a subset of phrases as the reduced subset of phrases for the advertisement. ### Claim 8 (depends on 6) The non-transitory computer-readable storage medium of claim 6, wherein the content of the advertisement comprises at least one of a title of the advertisement, a description of the advertisement, or a Uniform Resource Locator (URL). ### Claim 9 (depends on 6) The non-transitory computer-readable storage medium of claim 6, wherein analyzing the set of descriptions further comprises identifying a phrase with a Term Frequency-Inverse Document Frequency (TF-IDF) score for the second set of phrases above a threshold value, the second set of phrases being denoted as a query expansion feature set. ### Claim 10 (depends on 6) The non-transitory computer-readable storage medium of claim 6, wherein the combined Ad keywords and ad expanded feature set further comprises, normalizing a weighting value for each Ad keyword and phrase within the ad expanded feature set prior to combining. ### Claim 11 (depends on 6) The non-transitory computer-readable storage medium of claim 6, wherein the second content for display further comprises at least one of a web page or a search result. ### Claim 12 A computer-implemented method for managing content for delivery over a network, the method comprising: providing a networked device comprising a memory and a processor configured to execute instructions stored in the memory; receiving, by the processor, an advertisement with Ad keywords associated with the advertisement; segmenting, by the processor, content of the advertisement into a plurality of phrases; performing, by the processor, a semantic analysis to the plurality of phrases; selecting, by the processor, top-ranked phrases from the plurality of phrases based on the semantic analysis to form a reduced subset of phrases; generating, by the processor, a query list having a plurality of queries by combining each phrase in the reduced subset of phrases that has less than a defined number of words with another independent phrase in the reduced subset of phrases to create queries comprising two or more words; for each query within the query list, performing, by the processor, an internet web search to identify a set of descriptions, each description describing a search result from the internet search; selecting, by the processor, a second set of phrases denoted as a query expansion feature set for each query based on an analysis of the set of descriptions; ranking, by the processor, the second set of phrases; generating, by the processor, an expanded Ad index by selecting from the ranked second set of phrases a subset of phrases as an ad expanded feature set that is combined with the Ad keywords associated with the advertisement; and when receiving a second content for display at a client device, selecting, by the processor, an advertisement from a plurality of advertisements, including the received advertisement, based on the expanded Ad index. ### Claim 13 (depends on 12) The computer-implemented method of claim 12, wherein conducting the semantic analysis comprises: grouping related words from the content of the advertisement to generate a phrase from the content of the advertisement; determining a pointwise mutual information (PMI) value for each generated phrase with each Ad keyword; ranking in order the determined PMI values based on a Term Frequency-Inverse Document Frequency score for each generated phrase; and selecting from the rank ordering a subset of phrases as the reduced subset of phrases for the advertisement. ### Claim 14 (depends on 12) The computer-implemented method of claim 12, wherein the content of the advertisement comprises at least one of a title of the advertisement, a description of the advertisement, or a Uniform Resource Locator (URL). ### Claim 15 (depends on 12) The computer-implemented method of claim 12, wherein analyzing the set of descriptions comprises identifying a phrase with a Term Frequency-Inverse Document Frequency (TF-IDF) score for the second set of phrases above a threshold value, the selected other set of phrases being denoted as a query expansion feature set. ### Claim 16 (depends on 12) The computer-implemented method of claim 12, wherein combining the Ad keywords and ad expanded feature set comprises normalizing a weighting value for each Ad keyword and phrase within the ad expanded feature set prior to the combination. ### Claim 17 (depends on 12) The computer-implemented method of claim 12, wherein the client requested content further comprises a web page or a search result.