# US 10,769,532 B2 · Network rating prediction engine > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,769,532 B2 - Document ID: US10769532B2 - Status: Granted - Kind: B2 - Application number: 15/480,011 - Filed: 2017-04-05 - Published: 2018-10-11 - Granted: 2020-09-08 - Earliest claimed: 2017-04-05 - Assignees: Accenture Global Solutions Limited - Applicants: Accenture Global Solutions Limited - Inventors: Congwei Dang; Takuya Kudo; Takafumi Mizuno; Makoto Yomosa; Sayaka Tanaka; Miku Yoshio - Web page: https://patentgrep.com/patent/US10769532B2 ## CPC - G06N 3/088: Non-supervised learning, e.g. competitive learning - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections - G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning - G06N 3/09: Supervised learning - G06N 3/098: Distributed learning, e.g. federated learning - G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn - G06F 40/30: Semantic analysis - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A rating prediction engine builds and applies models to predict ratings based on an analysis of textual reviews and comments. The engine can build multiple models simultaneously through distributed parallel model building that employs deep convolutional neural networks (CNNs). The engine can also incorporate user moment feature data, including user status and context information, to provide better performance and more accurate predictions. The engine can also employ heuristic unsupervised pre-training and/or adaptive over-fitting reduction for model building. In some instances, the techniques described herein can be used in a service to predict personalized ratings for reviews or other published items, in instances where the original author of the item did not include a rating and/or in instances where the publication channel does not provide a mechanism to enter ratings. ## Claims ### Claim 1 A computer-implemented method performed by at least one processor, the method comprising: receiving, by the at least one processor, first review data that includes a plurality of rated reviews; determining, by the at least one processor, a plurality of model building transactions to be executed to build at least one prediction model based on the first review data; building, by the at least one processor, the at least one prediction model through use of at least one deep convolutional neural network (CNN), including distributing the plurality of model building transactions for parallel execution on a plurality of compute nodes, the building comprising a heuristic unsupervised pre-training that employs a heuristic formula that includes an absolute difference factor and a distribution difference factor; receiving, by the at least one processor, second review data that includes a plurality of unrated reviews; and predicting, by the at least one processor, a rating for each of the plurality of unrated reviews using the at least one prediction model. ### Claim 2 (depends on 1) The method of claim 1, wherein: the plurality of model building transactions are distributed, through a scheduler, based on load information for the plurality of compute nodes; and the load information is received by the scheduler from a plurality of training managers each executing on a respective compute node. ### Claim 3 (depends on 1) The method of claim 1, wherein the building of the at least one prediction model includes a user moment feature fusion to provide a plurality of unified features by combining a plurality of user moment features with a plurality of features that are output from the at least one deep CNN. ### Claim 4 (depends on 3) The method of claim 3, wherein the building of the at least one prediction model includes an adaptive over-fitting reduction to reduce the plurality of unified features. ### Claim 5 (depends on 4) The method of claim 4, wherein the adaptive over-fitting reduction includes, in one or more iterations: randomly dropping out a subset of the unified features; and determining a strength parameter based on a difference between outputs of a training iteration applied to training data and validation data, wherein the strength parameter indicates an amount of random dropout in a next iteration. ### Claim 6 (depends on 1) The method of claim 1, wherein the at least one deep CNN includes a plurality of convolution layers of varying scope. ### Claim 7 (depends on 6) The method of claim 6, wherein the plurality of convolution layers includes layers that correspond to a phrase feature map group, a sentence feature map group, a paragraph feature map group, and a context feature map group. ### Claim 8 A system comprising: at least one processor; and a memory communicatively coupled to the at least one processor, the memory storing instructions which, when executed by the at least one processor, instruct the at least one processor to perform operations comprising: receiving first review data that includes a plurality of rated reviews; determining a plurality of model building transactions to be executed to build at least one prediction model based on the first review data; building the at least one prediction model through use of at least one deep convolutional neural network (CNN), including distributing the plurality of model building transactions for parallel execution on a plurality of compute nodes, the building comprising a heuristic unsupervised pre-training that employs a heuristic formula that includes an absolute difference factor and a distribution difference factor; receiving second review data that includes a plurality of unrated reviews; and predicting a rating for each of the plurality of unrated reviews using the at least one prediction model. ### Claim 9 (depends on 8) The system of claim 8, wherein: the plurality of model building transactions are distributed, through a scheduler, based on load information for the plurality of compute nodes; and the load information is received by the scheduler from a plurality of training managers each executing on a respective compute node. ### Claim 10 (depends on 8) The system of claim 8, wherein the building of the at least one prediction model includes a user moment feature fusion to provide a plurality of unified features by combining a plurality of user moment features with a plurality of features that are output from the at least one deep CNN. ### Claim 11 (depends on 10) The system of claim 10, wherein the building of the at least one prediction model includes an adaptive over-fitting reduction to reduce the plurality of unified features. ### Claim 12 (depends on 11) The system of claim 11, wherein the adaptive over-fitting reduction includes, in one or more iterations: randomly dropping out a subset of the unified features; and determining a strength parameter based on a difference between outputs of a training iteration applied to training data and validation data, wherein the strength parameter indicates an amount of random dropout in a next iteration. ### Claim 13 (depends on 8) The system of claim 8, wherein the at least one deep CNN includes a plurality of convolution layers of varying scope. ### Claim 14 (depends on 13) The system of claim 13, wherein the plurality of convolution layers includes layers that correspond to a phrase feature map group, a sentence feature map group, a paragraph feature map group, and a context feature map group. ### Claim 15 One or more non-transitory computer-readable storage media storing instructions which, when executed by at least one processor, instruct the at least one processor to perform operations comprising: receiving first review data that includes a plurality of rated reviews; determining a plurality of model building transactions to be executed to build at least one prediction model based on the first review data; building the at least one prediction model through use of at least one deep convolutional neural network (CNN), including distributing the plurality of model building transactions for parallel execution on a plurality of compute nodes, the building comprising a heuristic unsupervised pre-training that employs a heuristic formula that includes an absolute difference factor and a distribution difference factor; receiving second review data that includes a plurality of unrated reviews; and predicting a rating for each of the plurality of unrated reviews using the at least one prediction model. ### Claim 16 (depends on 15) The one or more non-transitory computer-readable storage media of claim 15, wherein: the plurality of model building transactions are distributed, through a scheduler, based on load information for the plurality of compute nodes; and the load information is received by the scheduler from a plurality of training managers each executing on a respective compute node. ### Claim 17 (depends on 15) The one or more non-transitory computer-readable storage media of claim 15, wherein the building of the at least one prediction model includes a user moment feature fusion to provide a plurality of unified features by combining a plurality of user moment features with a plurality of features that are output from the at least one deep CNN.