# US 12,770,147 B2 · Multi-task convolutional neural network for behavior sequence embedding modeling > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,770,147 B2 - Document ID: US12770147B2 - Status: Granted - Kind: B2 - Application number: 18/679,129 - Filed: 2024-05-30 - Published: 2025-12-04 - Granted: 2026-10-20 - Earliest claimed: 2024-05-30 - Assignees: eBay Inc. - Applicants: eBay Inc. - Inventors: Minghao Gu; Bo Qu; Zhurong Wang; Daisuke Yagi; Zitao Zhang; Zhichao Han; Yang Zhao - Web page: https://patentgrep.com/patent/US12770147B2 ## CPC - G06Q 20/40145: {Biometric identity checks} - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06Q 50/265: {Personal security, identity or safety} - 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 Some aspects of the present technology relate to performing fraud detection on online transaction platforms through user behavior sequence data. A multi-task convolutional neural network (MTCNN) model is used to predict, in real-time, whether user behavior sequence data is indicative of fraudulent activity. To perform fraud detection in such configurations, a one-layer convolutional neural network architecture with multi-range kernels is employed. The MTCNN model receives a sequence of page browsing signals corresponding to a buyer. Each page browsing signal corresponds to a position in the sequence. One or more portions of the page browsing signals are selected. Each of the one or more portions of the page browsing signals and the corresponding position are embedded in one or more sequence embeddings. A fraud risk for each of the one or more sequence embeddings is predicted utilizing the MTCNN model. ## Claims ### Claim 1 One or more computer storage media storing computer-usable instructions that, when used by one or more computing devices, cause the one or more computing devices to perform operations, the operations comprising: accessing, by at least one server from one or more servers of an online transaction platform, a sequence of page browsing signals corresponding to user interactions of a user with the online transaction platform, each of the page browsing signals corresponding to a position in the sequence; selecting, by at least one server from the one or more servers, a plurality of temporally distinct portions of the sequence of page browsing signals, the plurality of temporally distinct portions comprising at least a first portion spanning a first temporal window of user interactions and a second portion spanning a second, longer temporal window of user interactions that overlaps the first temporal window; generating, by at least one server from the one or more servers, a plurality of sequence embeddings, wherein each sequence embedding is generated for a corresponding temporally distinct portion from the plurality of temporally distinct portions using the page browsing signals in the corresponding temporally distinct portion and positional information indicative of an order of the page browsing signals in the corresponding temporally distinct portion; predicting, by at least one server from the one or more servers utilizing a multi-task model with multi-range kernels trained to detect fraudulent activity in online transactions, a fraud risk for each of the plurality of sequence embeddings in parallel, wherein each multi-range kernel convolves over a different temporal span corresponding to a different one of the plurality of temporally distinct portions; and causing, by at least one server from the one or more servers, an action to be performed on the online transaction platform based on the fraud risk for at least one of the plurality of sequence embeddings, the action comprising at least one of blocking a transaction involving the user, freezing an account of the user, or closing the account of the user. ### Claim 2 (depends on 1) The one or more computer storage media of claim 1, further comprising identifying a purchase signal corresponding to the user, the purchase signal indicating the user has attempted to purchase an item. ### Claim 3 (depends on 2) The one or more computer storage media of claim 2, further comprising, upon identifying the purchase signal, requesting the sequence of page browsing signals. ### Claim 4 (depends on 1) The one or more computer storage media of claim 1, wherein the fraud risk for each of the plurality of sequence embeddings indicates the user is attempting transaction fraud for a current transaction, and the action comprises blocking the current transaction in real-time. ### Claim 5 (depends on 1) The one or more computer storage media of claim 1, wherein the multi-task model is a convolutional neural network trained to detect one or more types of fraud risk in parallel utilizing the multi-range kernels. ### Claim 6 (depends on 1) The one or more computer storage media of claim 1, wherein the page browsing signals comprise page identification, item identification, and view time. ### Claim 7 (depends on 1) The one or more computer storage media of claim 1, wherein the multi-task model is a one-layer convolutional neural network architecture. ### Claim 8 (depends on 1) The one or more computer storage media of claim 1, wherein the multi-task model is trained with random label weights. ### Claim 9 (depends on 1) The one or more computer storage media of claim 1, wherein the multi-task model is trained for each of one or more types of transaction fraud in parallel. ### Claim 10 (depends on 1) The one or more computer storage media of claim 1, wherein the fraud risk for each of the plurality of sequence embeddings is indicative of an account takeover, stolen financial information, or high risk buying. ### Claim 11 A computer-implemented method comprising: accessing, by at least one server from one or more servers of an online transaction platform, a sequence of page browsing signals corresponding to user interactions of a user with the online transaction platform, each of the page browsing signals corresponding to a position in the sequence; selecting, by at least one server from the one or more servers, a plurality of temporally distinct portions of the sequence of page browsing signals, the plurality of temporally distinct portions comprising at least a first portion spanning a first temporal window of user interactions and a second portion spanning a second, longer temporal window of user interactions that overlaps the first temporal window; generating, by at least one server from the one or more servers, a plurality of sequence embeddings, wherein each sequence embedding is generated for a corresponding temporally distinct portion from the plurality of temporally distinct portions using the page browsing signals in the corresponding temporally distinct portion and positional information indicative of an order of the page browsing signals in the corresponding temporally distinct portion; predicting, by at least one server from the one or more servers utilizing a multi-task model with multi-range kernels trained to detect fraudulent activity in online transactions, a fraud risk for each of the plurality of sequence embeddings in parallel, wherein each multi-range kernel convolves over a different temporal span corresponding to a different one of the plurality of temporally distinct portions; and causing, by at least one server from the one or more servers, an action to be performed on the online transaction platform based on the fraud risk for at least one of the plurality of sequence embeddings, the action comprising at least one of blocking a transaction involving the user, freezing an account of the user, or closing the account of the user. ### Claim 12 (depends on 11) The computer-implemented method of claim 11, further comprising identifying a purchase signal corresponding to the user, the purchase signal indicating the user has attempted to purchase an item. ### Claim 13 (depends on 12) The computer-implemented method of claim 12, further comprising, upon identifying the purchase signal, requesting the sequence of page browsing signals. ### Claim 14 (depends on 11) The computer-implemented method of claim 11, wherein the fraud risk for each of the plurality of sequence embeddings indicates the user is attempting transaction fraud for a current transaction, and the action comprises blocking the current transaction in real-time. ### Claim 15 (depends on 11) The computer-implemented method of claim 11, wherein the multi-task model is a convolutional neural network trained to detect one or more types of transaction fraud in parallel utilizing the multi-range kernels. ### Claim 16 A computer system comprising: one or more processors; and one or more computer storage medium storing computer-usable instructions that, when used by the one or more processors, causes the computer system to perform operations comprising: accessing, by at least one server from one or more servers of an online transaction platform, a sequence of page browsing signals corresponding to user interactions of a user with the online transaction platform, each of the page browsing signals corresponding to a position in the sequence; selecting, by at least one server from the one or more servers, a plurality of temporally distinct portions of the sequence of page browsing signals, the plurality of temporally distinct portions comprising at least a first portion spanning a first temporal window of user interactions and a second portion spanning a second, longer temporal window of user interactions that overlaps the first temporal window; generating, by at least one server from the one or more servers, a plurality of sequence embeddings, wherein each sequence embedding is generated for a corresponding temporally distinct portion from the plurality of temporally distinct portions using the page browsing signals in the corresponding temporally distinct portion and positional information indicative of an order of the page browsing signals in the corresponding temporally distinct portion; predicting, by at least one server from the one or more servers utilizing a multi-task model with multi-range kernels trained to detect fraudulent activity in online transactions, a fraud risk for each of the plurality of sequence embeddings in parallel, wherein each multi-range kernel convolves over a different temporal span corresponding to a different one of the plurality of temporally distinct portions; and causing, by at least one server from the one or more servers, an action to be performed on the online transaction platform based on the fraud risk for at least one of the plurality of sequence embeddings, the action comprising at least one of blocking a transaction involving the user, freezing an account of the user, or closing the account of the user. ### Claim 17 (depends on 16) The computer system of claim 16, further comprising identifying a purchase signal corresponding to the user, the purchase signal indicating the user has attempted to purchase an item. ### Claim 18 (depends on 17) The computer system of claim 17, further comprising, upon identifying the purchase signal, requesting the sequence of page browsing signals. ### Claim 19 (depends on 16) The computer system of claim 16, wherein the fraud risk for each of the plurality of sequence embeddings indicates the user is attempting transaction fraud for a current transaction, and the action comprises blocking the current transaction in real-time. ### Claim 20 (depends on 16) The computer system of claim 16, wherein the multi-task model is a convolutional neural network trained to detect one or more types of transaction fraud in parallel utilizing the multi-range kernels.