# US 11,443,347 B2 · System and method for click-through rate prediction > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,347 B2 - Document ID: US11443347B2 - Status: Granted - Kind: B2 - Application number: 17/006,495 - Filed: 2020-08-28 - Published: 2021-03-04 - Granted: 2022-09-13 - Earliest claimed: 2019-08-30 - Assignees: Samsung Electronics Co., Ltd. - Applicants: Samsung Electronics Co., Ltd. - Inventors: Xianshan Qu; Li Li; Xi Liu; Rui Chen; Yong Ge; Soo-Hyun Choi - Web page: https://patentgrep.com/patent/US11443347B2 ## CPC - G06Q 30/0275: {Auctions} - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/0499: Feedforward networks - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - G06Q 30/0244: {Optimization} - G06Q 30/0271: {Personalized advertisement} - G06Q 30/0276: {Advertisement creation} - G06N 3/045: Combinations of networks - 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 - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A system and method capable of learning dynamic user and advertisement behavior for more effective click-through rate prediction. The system and method include at least one processor configured to obtain at least one item data, wherein the at least one item data comprises at least one explicit feedback to a user interaction event and other data associated with an item. The at least one processor also uses an interaction model that incorporates the obtained at least one item data to generate a user response prediction for a user and another interaction event. ## Claims ### Claim 1 A method comprising: obtaining, by an electronic device, at least one item data, wherein the at least one item data comprises (i) at least one explicit feedback to a user interaction event and (ii) historical advertisement data comprising one or more bid requests; identifying, by the electronic device, at least one external data source; aligning, by the electronic device, the at least one item data to the at least one external data source based on at least one contextual data, creating pairs of aligned data, wherein for each of the pairs of the aligned data, aligning the at least one item data comprises: computing a weighted linear combination of a temporal interest of a user and a previous temporal interest of the user; and passing the weighted linear combination as an input to a next cell of an interaction model; generating, by the electronic device, an aggregate result based on passing the weighted linear combinations through a series of cells of the interaction model; and generating, by the electronic device using the interaction model, a user response prediction for the user and another interaction event. ### Claim 2 (depends on 1) The method of claim 1, further comprising: obtaining, by the electronic device, at least one user feature and at least one item feature; and identifying, by the electronic device, at least one user dynamic feature of the at least one user feature and at least one item dynamic feature of the at least one item feature. ### Claim 3 (depends on 2) The method of claim 2, further comprising: applying, by the electronic device, a first feedforward neural network to the at least one user dynamic feature; and applying, by the electronic device, a second feedforward neural network to the at least one item dynamic feature. ### Claim 4 (depends on 2) The method of claim 2, further comprising: obtaining, by the electronic device, the at least one user dynamic feature and the at least one item dynamic feature; determining, by the electronic device, an aggregate result of the at least one user dynamic feature based on historical data associated with the at least one user dynamic feature; and determining, by the electronic device, an aggregate result of the at least one item dynamic feature based on historical data associated with the at least one item dynamic feature. ### Claim 5 (depends on 1) The method of claim 1, wherein the item data further comprises: historical user data comprising a series of user activities with a temporal dependency, wherein the user interaction event further comprises an interaction result in response to a user interaction with an advertisement, and wherein the user response prediction further comprises predicting in real-time, by the electronic device, a click-through rate for the user in response to an incoming advertisement bid request. ### Claim 6 (depends on 1) The method of claim 1 further comprising: in response to receiving an advertisement bid request from an ad exchange server, determining, by the electronic device, at least one predicted click through rate value; computing, by the electronic device, at least one bid price based on the predicted click through rate value; determining, by the electronic device, a winning advertisement; and outputting, by the electronic device, the winning advertisement. ### Claim 7 An electronic device comprising: at least one memory; and at least one processor operatively coupled to the at least one memory, the at least one processor configured to: obtain at least one item data, wherein the at least one item data comprises (i) at least one explicit feedback to a user interaction event and (ii) historical advertisement data comprising one or more bid requests; identify at least one external data source; align the at least one item data to the at least one external data source based on at least one contextual data, creating pairs of aligned data, wherein for each of the pairs of the aligned data, the at least one processor is configured to: compute a weighted linear combination of a temporal interest of a user and a previous temporal interest of the user; and pass the weighted linear combination as an input to a next cell of an interaction model; generate an aggregate result based on passing the weighted linear combinations through a series of cells of the interaction model; and generate, using the interaction model, a user response prediction for the user and another interaction event. ### Claim 8 (depends on 7) The electronic device of claim 7, wherein the at least one processor is further configured to: obtain at least one user feature and at least one item feature; and identify at least one user dynamic feature of the at least one user feature and at least one item dynamic feature of the at least one item feature. ### Claim 9 (depends on 8) The electronic device of claim 8, wherein the at least one processor is further configured to: apply a first feedforward neural network to the at least one user dynamic feature; and apply a second feedforward neural network to the at least one item dynamic feature. ### Claim 10 (depends on 8) The electronic device of claim 8, wherein the at least one processor is further configured to: obtain the at least one user dynamic feature and the at least one item dynamic feature; determine an aggregate result of the at least one user dynamic feature based on historical data associated with the at least one user dynamic feature; and determine an aggregate result of the at least one item dynamic feature based on historical data associated with the at least one item dynamic feature. ### Claim 11 (depends on 7) The electronic device of claim 7, wherein the item data further comprises: historical user data comprising a series of user activities with a temporal dependency, wherein the user interaction event further comprises an interaction result in response to a user interaction with an advertisement, and wherein the user response prediction further comprises predicting in real-time, by the electronic device, a click-through rate for the user in response to an incoming advertisement bid request. ### Claim 12 (depends on 7) The electronic device of claim 7, wherein the at least one processor is further configured to: in response to receiving an advertisement bid request from an ad exchange server, determine at least one predicted click through rate value; compute at least one bid price based on the predicted click through rate value; determine a winning advertisement; and output the winning advertisement. ### Claim 13 A non-transitory machine-readable medium comprising instructions that, when executed by at least one processor of an electronic device, cause the at least one processor to: obtain at least one item data, wherein the at least one item data comprises (i) at least one explicit feedback to a user interaction event and (ii) historical advertisement data comprising one or more bid requests; identify at least one external data source; align the at least one item data to the at least one external data source based on at least one contextual data, creating pairs of aligned data, wherein for each of the pairs of the aligned data, the at least one processor is configured to: compute a weighted linear combination of a temporal interest of a user and a previous temporal interest of the user; and pass the weighted linear combination as an input to a next cell of an interaction model; generate an aggregate result based on passing the weighted linear combinations through a series of cells of the interaction model; and generate, using the interaction model, a user response prediction for the user and another interaction event. ### Claim 14 (depends on 13) The non-transitory machine-readable medium of claim 13, wherein the instructions are further configured to cause the at least one processor to: obtain at least one user feature and at least one item feature; and identify at least one user dynamic feature of the at least one user feature and at least one item dynamic feature of the at least one item feature. ### Claim 15 (depends on 14) The non-transitory machine-readable medium of claim 14, wherein the instructions are further configured to cause the at least one processor to: apply a first feedforward neural network to the at least one user dynamic feature; and apply a second feedforward neural network to the at least one item dynamic feature. ### Claim 16 (depends on 13) The non-transitory machine-readable medium of claim 13, wherein the item data further comprises: historical user data comprising a series of user activities with a temporal dependency, wherein the user interaction event further comprises an interaction result in response to a user interaction with an advertisement, and wherein the user response prediction further comprises predicting in real-time, by the electronic device, a click-through rate for the user in response to an incoming advertisement bid request. ### Claim 17 (depends on 13) The non-transitory machine-readable medium of claim 13, wherein the instructions are further configured to cause the at least one processor to: in response to receiving an advertisement bid request from an ad exchange server, determine at least one predicted click through rate value; compute at least one bid price based on the predicted click through rate value; determine a winning advertisement; and output the winning advertisement. ### Claim 18 (depends on 1) The method of claim 1, wherein the interaction model comprises at least one of: a long-short term memory (LSTM) model and a contextual long-short term memory (CLSTM) model. ### Claim 19 (depends on 7) The electronic device of claim 7, wherein the interaction model comprises at least one of: a long-short term memory (LSTM) model and a contextual long-short term memory (CLSTM) model. ### Claim 20 (depends on 13) The non-transitory machine-readable medium of claim 13, wherein the interaction model comprises at least one of: a long-short term memory (LSTM) model and a contextual long-short term memory (CLSTM) model.