Method and device for identifying the falsehood of information

Abstract

The present invention discloses a method and device for identifying the degree of falsehood of information, and relates to the field of computer technology. A specific implementation of the method includes: receiving an identification request, obtaining information to be identified of a user's comments on a product according to the identification request, and converting the information to be identified into a heterogeneous graph to be identified; calling a feature calculation method set to calculate the user's initial features, comment initial features, and product initial features in the information to be identified; inputting the heterogeneous graph to be identified, the user's initial features, the comment initial features, and the product initial features into a trained neural network model to obtain an identification result of the degree of falsehood of the information to be identified; wherein the neural network model is trained based on historical information of user comments on products with known falsehood levels. This implementation can convert the information of user comments on products into a heterogeneous graph, and then use a graph neural network model to effectively fuse the graph structure information and the node features in the graph for detection to determine the falsehood degree of the comments.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for identifying the falsehood of information. Background Art

[0002] In the era of big data, the need for data anomaly detection is common. Anomaly detection uses computer data mining technology to find abnormal behaviors or behavior sequences that are inconsistent with normal data. Anomaly detection technology can be applied to network security intrusion detection, e-commerce platform comment detection, Internet water army detection, fake news detection and other fields. Common anomaly detection methods include (1) detection based on graph structure models and (2) detection based on graph neural network models.

[0003] In the process of implementing the present invention, there are at least the following problems in the prior art:

[0004] (1) For detection based on graph structure models, only the structural information of the graph is often considered without integrating the feature information of the node itself;

[0005] (2) For detection based on graph neural network models, current models are mostly applicable to isomorphic graphs and do not consider whether adjacent nodes are disguised nodes, resulting in inaccurate detection results. Summary of the Invention

[0006] In view of this, an embodiment of the present invention provides a method and device for identifying the degree of falsehood of information, which can convert the information to be identified in user reviews of products into a heterogeneous graph, and then use a graph neural network model to effectively fuse the graph structure information and the features of the nodes in the graph for detection to determine the degree of falsehood of the reviews.

[0007] To achieve the above object, according to one aspect of an embodiment of the present invention, a method for identifying the degree of falsity of information is provided, comprising:

[0008] Receiving a recognition request, obtaining information to be recognized of a user's comments on a product according to the recognition request, and converting the information to be recognized into a heterogeneous graph to be recognized;

[0009] Retrieving a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified;

[0010] The heterogeneous graph to be identified, the initial features of the user, the initial features of the comment, and the initial features of the product are input into a trained neural network model to obtain a falsehood degree identification result of the information to be identified; wherein the neural network model is trained based on historical information of user comments on products with known falsehood degrees.

[0011] Optionally, before receiving the identification request, the method further includes:

[0012] Generate the identification request for all the review information in a preset time period according to a preset frequency; or

[0013] The identification request is generated based on the received identification dimension information.

[0014] Optionally, converting the information to be identified into a heterogeneous graph to be identified includes:

[0015] Determine the user node, comment node, and product node in the heterogeneous graph to be identified based on the user information, comment content, and product information in the information to be identified; wherein the comment node is the node to be identified, and the user node and the product node are adjacent nodes of the comment node;

[0016] Determine, based on the association relationship between the user, the comment, and the product indicated by the user's comment on the product in the information to be identified, the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified;

[0017] Based on the user nodes, comment nodes, product nodes, edges between user nodes and comment nodes, and edges between comment nodes and product nodes, the heterogeneous graph to be identified is constructed.

[0018] Optionally, the neural network model is trained according to the following method, including:

[0019] Retrieve historical information about product reviews from users with known falsehood levels, and select a preset amount of information from them as training data;

[0020] The training data is converted into a sample heterogeneous graph containing user nodes, comment nodes, and product nodes as training samples; wherein the comment node is the node to be identified, and the user node and the product node are both adjacent nodes of the comment node; and the falsehood degree of historical information in the training data is used as the sample label of the node to be identified;

[0021] Retrieving a feature calculation method set to calculate user initial features, comment initial features, and product initial features in the training data as sample features;

[0022] The neural network model is obtained by training according to the training samples, sample labels, and sample features.

[0023] Optionally, the training of the neural network model according to the training samples, sample labels, and sample features includes:

[0024] For each comment node in the sample heterogeneous graph, determine its adjacent user nodes and adjacent product nodes;

[0025] For the adjacent user node, determine other review nodes and other product nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial product features, call an aggregation function to generate a first user representation vector for the adjacent user node; concatenate the first user representation vector with the initial user features of the adjacent user node to obtain a user representation vector for the adjacent user node;

[0026] For the adjacent product node, determine other review nodes and other user nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial user features, and call an aggregation function to generate a first product representation vector for the adjacent product node; concatenate the first product representation vector with the initial product features of the adjacent product node to obtain a product representation vector for the adjacent product node;

[0027] The sample heterogeneous graph, user representation vector, product representation vector, initial comment features of comment nodes, and the sample labels are connected to a multi-layer neural network and trained to obtain the neural network model.

[0028] Optionally, calling an aggregation function to generate a first user representation vector of the adjacent user node, or calling an aggregation function to generate a first product representation vector of the adjacent product node, includes:

[0029] An attention mechanism is used in the aggregation function to generate the first user representation vector or the first product representation vector.

[0030] Optionally, for each adjacent user node of a comment node, after determining other user nodes that can be reached by the adjacent product node of the comment node in the sample heterogeneous graph, the user similarity between the adjacent user node and the other user nodes is calculated based on the initial user features, and a preset filtering method is used to determine a second user node from the other user nodes based on the user similarity, and to determine a corresponding second product node; based on the second user node and the corresponding second product node, a user representation vector of the adjacent user node is determined; and / or,

[0031] For each adjacent product node of a comment node, after determining other product nodes that can be reached by the adjacent user node of the comment node in the sample heterogeneous graph, the product similarity between the adjacent product node and the other product nodes is calculated based on the initial product features, and a preset filtering method is used to determine a third product node from the other product nodes based on the product similarity, and determine a corresponding third user node; based on the third product node and the corresponding third user node, the product representation vector of the adjacent product node is determined.

[0032] Optionally, the user initial feature includes at least one of the following: a feature of the number of comments posted by the user, a feature of the level of comments posted by the user, a feature of the frequency of comments posted by the user, and a feature of the length of comments posted by the user;

[0033] The initial review feature includes at least one of the following: a grade feature, a grade deviation feature, a content feature, and a feature of the corresponding product being reviewed;

[0034] The initial product features include at least one of the following: a feature of the number of reviews, a feature of the review level, a feature of the review content, and a feature of the review frequency.

[0035] According to another aspect of an embodiment of the present invention, a device for identifying the degree of falsity of information is provided, comprising:

[0036] An information acquisition module is configured to receive an identification request, obtain information to be identified of a user's comments on a product according to the identification request, and convert the information to be identified into a heterogeneous graph to be identified;

[0037] A feature calculation module is used to call a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified;

[0038] The identification module is used to input the heterogeneous graph to be identified, the initial features of the user, the initial features of the comment, and the initial features of the product into a trained neural network model to obtain a falsehood degree identification result of the information to be identified; wherein, the neural network model is trained based on historical information of user comments on products with known falsehood degrees.

[0039] Optionally, before receiving the identification request, the method further includes:

[0040] Generate the identification request for all the review information in a preset time period according to a preset frequency; or

[0041] The identification request is generated based on the received identification dimension information.

[0042] Optionally, converting the information to be identified into a heterogeneous graph to be identified includes:

[0043] Determine the user node, comment node, and product node in the heterogeneous graph to be identified based on the user information, comment content, and product information in the information to be identified; wherein the comment node is the node to be identified, and the user node and the product node are adjacent nodes of the comment node;

[0044] Determine, based on the association relationship between the user, the comment, and the product indicated by the user's comment on the product in the information to be identified, the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified;

[0045] Based on the user nodes, comment nodes, product nodes, edges between user nodes and comment nodes, and edges between comment nodes and product nodes, the heterogeneous graph to be identified is constructed.

[0046] Optionally, the neural network model is trained according to the following method, including:

[0047] Retrieve historical information about product reviews from users with known falsehood levels, and select a preset amount of information from them as training data;

[0048] The training data is converted into a sample heterogeneous graph containing user nodes, comment nodes, and product nodes as training samples; wherein the comment node is the node to be identified, and the user node and the product node are both adjacent nodes of the comment node; and the falsehood degree of historical information in the training data is used as the sample label of the node to be identified;

[0049] Retrieving a feature calculation method set to calculate user initial features, comment initial features, and product initial features in the training data as sample features;

[0050] The neural network model is obtained by training according to the training samples, sample labels, and sample features.

[0051] Optionally, the training of the neural network model according to the training samples, sample labels, and sample features includes:

[0052] For each comment node in the sample heterogeneous graph, determine its adjacent user nodes and adjacent product nodes;

[0053] For the adjacent user node, determine other review nodes and other product nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial product features, call an aggregation function to generate a first user representation vector for the adjacent user node; concatenate the first user representation vector with the initial user features of the adjacent user node to obtain a user representation vector for the adjacent user node;

[0054] For the adjacent product node, determine other review nodes and other user nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial user features, and call an aggregation function to generate a first product representation vector for the adjacent product node; concatenate the first product representation vector with the initial product features of the adjacent product node to obtain a product representation vector for the adjacent product node;

[0055] The sample heterogeneous graph, user representation vector, product representation vector, initial comment features of comment nodes, and the sample labels are connected to a multi-layer neural network and trained to obtain the neural network model.

[0056] Optionally, calling an aggregation function to generate a first user representation vector of the adjacent user node, or calling an aggregation function to generate a first product representation vector of the adjacent product node, includes:

[0057] An attention mechanism is used in the aggregation function to generate the first user representation vector or the first product representation vector.

[0058] Optionally, for each adjacent user node of a comment node, after determining other user nodes that can be reached by the adjacent product node of the comment node in the sample heterogeneous graph, the user similarity between the adjacent user node and the other user nodes is calculated based on the initial user features, and a preset filtering method is used to determine a second user node from the other user nodes based on the user similarity, and to determine a corresponding second product node; based on the second user node and the corresponding second product node, a user representation vector of the adjacent user node is determined; and / or,

[0059] For each adjacent product node of a comment node, after determining other product nodes that can be reached by the adjacent user node of the comment node in the sample heterogeneous graph, the product similarity between the adjacent product node and the other product nodes is calculated based on the initial product features, and a preset filtering method is used to determine a third product node from the other product nodes based on the product similarity, and determine a corresponding third user node; based on the third product node and the corresponding third user node, the product representation vector of the adjacent product node is determined.

[0060] Optionally, the user initial feature includes at least one of the following: a feature of the number of comments posted by the user, a feature of the level of comments posted by the user, a feature of the frequency of comments posted by the user, and a feature of the length of comments posted by the user;

[0061] The initial review feature includes at least one of the following: a grade feature, a grade deviation feature, a content feature, and a feature of the corresponding product being reviewed;

[0062] The initial product features include at least one of the following: a feature of the number of reviews, a feature of the review level, a feature of the review content, and a feature of the review frequency.

[0063] According to another aspect of an embodiment of the present invention, there is provided an electronic device for identifying the degree of falsity of information, comprising:

[0064] one or more processors;

[0065] a storage device for storing one or more programs,

[0066] When the one or more programs are executed by the one or more processors, the one or more processors implement the method for identifying the degree of falsehood of information provided by the present invention.

[0067] According to another aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the method for identifying the degree of falsity of information provided by the present invention is implemented.

[0068] One embodiment of the above invention has the following advantages or beneficial effects: because it adopts a technical means of converting the information to be identified of user comments on products into a heterogeneous graph, and then using a graph neural network model to effectively fuse the graph structure information and the characteristics of the nodes in the graph for detection to determine the degree of falsehood of the comments, it overcomes the technical problem that the graph neural network models in the prior art are mostly applicable to homogeneous graphs or do not consider the characteristics of the graph nodes themselves, and are therefore not applicable to the comment detection scenario in the present invention, thereby achieving the technical effect of being able to use the graph neural network model combined with the characteristics of the comment information itself to detect false comments.

[0069] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The accompanying drawings are provided for a better understanding of the present invention and are not intended to limit the present invention.

[0071] Figure 1 1 is a schematic diagram of the main process of a method for identifying the falsehood level of information according to the first embodiment of the present invention;

[0072] Figure 2 is a schematic diagram of a heterogeneous graph in a method for identifying the degree of false information according to a second embodiment of the present invention;

[0073] Figure 3 2. It is a schematic diagram of the transformation of a heterogeneous graph when training a neural network in a method for identifying the degree of false information according to the second embodiment of the present invention;

[0074] Figure 4 is a schematic diagram of main modules of a device for identifying the falsehood level of information according to an embodiment of the present invention;

[0075] Figure 5 is an exemplary system architecture diagram in which embodiments of the present invention may be applied;

[0076] Figure 6 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The following description of exemplary embodiments of the present invention is made in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, it should be appreciated by those skilled in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0078] Figure 1 FIG. 1 is a schematic diagram of the main process of a method for identifying the falsehood of information according to the first embodiment of the present invention. Figure 1 Shown, including:

[0079] Step S101: receiving a recognition request, obtaining information to be recognized of a user's comments on a product according to the recognition request, and converting the information to be recognized into a heterogeneous graph to be recognized;

[0080] Step S102: Retrieve a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified;

[0081] Step S103: Input the heterogeneous graph to be identified, the initial features of the user, the initial features of the comment, and the initial features of the product into a trained neural network model to obtain a falsehood degree identification result of the information to be identified; wherein, the neural network model is trained based on historical information of user comments on products with known falsehood degrees.

[0082] The scenario in this application can be a user commenting on a product (such as goods / catering / services, etc.). In actual situations, in order to attract more users, some businesses often hire water armies to give false positive reviews to their products; at the same time, competitors may also hire water armies to give false negative reviews to specific products, thereby interfering with users' normal judgment. These false reviews can be considered abnormal reviews relative to normal user reviews. The anomaly detection algorithm is to identify these abnormal samples from normal samples. The anomaly detection method provided in this application can be applied to the above-mentioned false review detection, and can also be applied to various application scenarios such as credit card fraud, network security intrusion detection, and medical image anomaly detection.

[0083] With the rise and application of deep learning, graph neural network models are widely used in anomaly detection. Embodiments of the present invention provide a method and device for identifying the degree of information falsity. These methods convert the to-be-identified information of user product reviews into a heterogeneous graph, and then use a graph neural network model to effectively integrate the graph structure information with the characteristics of the nodes in the graph to detect and determine the degree of falsity of the reviews.

[0084] In some embodiments, before receiving the identification request, the method further includes:

[0085] Generate the identification request for all the review information in a preset time period according to a preset frequency; or

[0086] The identification request is generated based on the received identification dimension information.

[0087] The identification request in this application can be generated based on a preset frequency set in advance (such as generating a request once a week, or generating a request once every half a month, etc.), or it can be generated when an external input identification dimension is received; in some practical applications, the identification dimension may include user level, product category, time range and other information. The identification dimension can be input by the user, or it can be an identification dimension indicated in the identification instruction issued by other upstream systems; it can be flexibly set according to actual conditions.

[0088] In some embodiments, converting the information to be identified into a heterogeneous graph to be identified includes:

[0089] Determine the user node, comment node, and product node in the heterogeneous graph to be identified based on the user information, comment content, and product information in the information to be identified; wherein the comment node is the node to be identified, and the user node and the product node are adjacent nodes of the comment node;

[0090] Determine, based on the association relationship between the user, the comment, and the product indicated by the user's comment on the product in the information to be identified, the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified;

[0091] Based on the user nodes, comment nodes, product nodes, edges between user nodes and comment nodes, and edges between comment nodes and product nodes, the heterogeneous graph to be identified is constructed.

[0092] The heterogeneous graph constructed according to the above method can present the information to be identified in a specific and complete manner in the form of a heterogeneous graph without missing the association relationship between nodes, and can be easily combined with node features.

[0093] In some embodiments, the neural network model is trained according to the following method, including:

[0094] Retrieve historical information about product reviews from users with known falsehood levels, and select a preset amount of information from them as training data;

[0095] The training data is converted into a sample heterogeneous graph containing user nodes, comment nodes, and product nodes as training samples; wherein the comment node is the node to be identified, and the user node and the product node are both adjacent nodes of the comment node; and the falsehood degree of historical information in the training data is used as the sample label of the node to be identified;

[0096] Retrieving a feature calculation method set to calculate user initial features, comment initial features, and product initial features in the training data as sample features;

[0097] The neural network model is obtained by training according to the training samples, sample labels, and sample features.

[0098] In the historical information of product reviews by users with a known degree of falsehood, each review may include the following content information: user-product-review content-rating-review time-whether it is a false review; among them, 0 can be used to indicate "true review" and 1 can be used to indicate "false review".

[0099] Figure 2 is a schematic diagram of a heterogeneous graph in a method for identifying the falsehood of information according to the second embodiment of the present invention, such as Figure 2 As shown, for a certain amount of selected training data, the converted heterogeneous graph can contain:

[0100] User node u i , product node p j , comment node r k , and user u i Post a comment k Form a type of edge In the product p j Comments on k Another type of edge

[0101] In some embodiments, the training of the neural network model based on the training samples, sample labels, and sample features includes:

[0102] For each comment node in the sample heterogeneous graph, determine its adjacent user nodes and adjacent product nodes;

[0103] For the adjacent user node, determine other review nodes and other product nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial product features, call an aggregation function to generate a first user representation vector for the adjacent user node; concatenate the first user representation vector with the initial user features of the adjacent user node to obtain a user representation vector for the adjacent user node;

[0104] For the adjacent product node, determine other review nodes and other user nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial user features, and call an aggregation function to generate a first product representation vector for the adjacent product node; concatenate the first product representation vector with the initial product features of the adjacent product node to obtain a product representation vector for the adjacent product node;

[0105] The sample heterogeneous graph, user representation vector, product representation vector, initial comment features of comment nodes, and the sample labels are connected to a multi-layer neural network and trained to obtain the neural network model.

[0106] Figure 3 : is a schematic diagram of the transformation of a heterogeneous graph when training a neural network in a method for identifying the falsehood of information according to the second embodiment of the present invention, such as Figure 3 As shown, for Figure 2 The transformation of r0 and r1 nodes in the heterogeneous graph includes:

[0107] For the r0 node, according to Figure 2 In heterogeneous graphs, determine the adjacent nodes u0 and p0 of node r0; then, for node u0, determine Figure 2 The comment nodes and corresponding product nodes that can be reached in the heterogeneous graph are: u0-r1-p1, u0-r6-p2; for the p0 node, determine Figure 2 The comment nodes and corresponding user nodes that can be reached in the heterogeneous graph: Figure 2 In the heterogeneous graph, it is determined to be none; Figure 3 As shown in the upper half of the picture.

[0108] Similarly, the same transformation method is used to transform the r1 node, such as Figure 3 As shown in the lower half of the picture.

[0109] In some practical applications, in order to facilitate computer processing, six connection matrices describing the sample heterogeneous graph can be generated according to the sample heterogeneous graph, and then calculations can be performed based on the six connection matrices according to computing requirements to obtain corresponding features or representation vectors.

[0110] Combined with the above Figure 2 , the six types of connection matrices can include:

[0111] (1) Connection matrix RU from comments to users k，i And the connection matrix RP from review to product k，j ; The kth row represents the comment r k is user u i For product p j Published.

[0112] exist Figure 2 The corresponding RU k，i is [[0][0][1][2][2][3][0][3]], the corresponding RP k，j is [[0][1][1][1][2][1][2][2]];

[0113] (2) User-to-comment matrix UR i，k and the user-to-product matrix UP i，j ,; where the i-th row represents user u i In the product p j Post a comment on k .exist Figure 2 The corresponding UR i，k For [[0, 1, 6][2][3, 4][5, 7]], the corresponding UP i，j is [[0, 1, 2][1][1, 2][1, 2]];

[0114] (3) Product-to-User Matrix PU j，i and product-to-review matrix PR j，k ; The jth row represents product p j Corresponding user u i and comments k .exist Figure 2 The corresponding PU j，i is [[0][0, 1, 2, 3][0, 2, 3]], the corresponding PR j，k is [[0][1, 2, 3, 5][4, 6, 7]].

[0115] Furthermore, when Figure 3 When transforming the heterogeneous graph as shown, we can first determine the comment r k , label 1 represents false comments and label 0 represents normal comments. k，i and RP k，j Find the corresponding user and product. i，k and UP i，j Find the comments and products corresponding to these users. j，i and PU j，k Find the corresponding comments and users, form Figure 3 The transformed data is shown in order to perform feature concatenation and obtain each representation vector.

[0116] The user representation vector, product representation vector, and initial comment features of the comment node obtained by the above method can fully and detailedly express the meaning of each node without missing the association relationship between nodes or the characteristic information of the node itself, thereby making the trained model more accurate.

[0117] In some embodiments, calling an aggregation function to generate a first user representation vector of the adjacent user node, or calling an aggregation function to generate a first product representation vector of the adjacent product node, includes:

[0118] An attention mechanism is used in the aggregation function to generate the first user representation vector or the first product representation vector.

[0119] Using the attention mechanism enables aggregated features to more accurately describe reviews. This translates to the following: when a user reviews many products, reviews that are closer to the product currently being reviewed are more valuable for consideration. For example, if a user has reviewed sports, food, and electronics, and their current review is about food, the attention mechanism can aggregate more of the user's food reviews while ignoring irrelevant information. This allows the aggregated features to more accurately describe the reviews, leading to a more accurate trained model.

[0120] In some embodiments, for each adjacent user node of a comment node, after determining other user nodes that can be reached by the adjacent product node of the comment node in the sample heterogeneous graph, the user similarity between the adjacent user node and the other user nodes is calculated based on the initial user features, and a preset filtering method is used to determine a second user node from the other user nodes based on the user similarity, and determine a corresponding second product node; based on the second user node and the corresponding second product node, a user representation vector of the adjacent user node is determined; and / or,

[0121] For each adjacent product node of a comment node, after determining other product nodes that can be reached by the adjacent user node of the comment node in the sample heterogeneous graph, the product similarity between the adjacent product node and the other product nodes is calculated based on the initial product features, and a preset filtering method is used to determine a third product node from the other product nodes based on the product similarity, and determine a corresponding third user node; based on the third product node and the corresponding third user node, the product representation vector of the adjacent product node is determined.

[0122] The above filtering method can include setting a filtering threshold or a percentage threshold; the following examples illustrate:

[0123] When the user similarity between an adjacent user node and other user nodes is calculated to be: 1, 0.9, 0.8, 0.7, 0.2, if the filtering method is to set the filtering threshold, then when the filtering threshold is 0.6, the user nodes corresponding to 1, 0.9, 0.8, 0.7 can be taken for subsequent calculations; if the filtering method is to set the filtering percentage threshold, then when the filtering threshold is 60%, the user nodes corresponding to 1, 0.9, 0.8 can be taken for subsequent calculations; under different scenario requirements, different filtering methods can be used and different filtering parameters (such as filtering threshold or filtering percentage threshold) can be set.

[0124] Using the above method, by screening out nodes that are more similar to the node of interest to determine the representation vector, it is possible to aggregate more effective node information and express node features more accurately, thereby improving model performance.

[0125] In some embodiments, the user initial feature includes at least one of the following: a feature of the number of comments posted by the user, a feature of the level of comments posted by the user, a feature of the frequency of comments posted by the user, and a feature of the length of comments posted by the user;

[0126] The initial review feature includes at least one of the following: a grade feature, a grade deviation feature, a content feature, and a feature of the corresponding product being reviewed;

[0127] The initial product features include at least one of the following: a feature of the number of reviews, a feature of the review level, a feature of the review content, and a feature of the review frequency.

[0128] In practical applications, the calculation method of each feature can be flexibly set according to actual conditions, or updated according to actual needs, so as to select more appropriate features in different scenarios and thus improve model performance.

[0129] Here are some example features:

[0130] Table 1 User characteristics

[0131]

[0132] Table 2 Product characteristics

[0133]

[0134]

[0135] Table 3 Review characteristics

[0136]

[0137]

[0138] Figure 4: is a schematic diagram of the main modules of a device for identifying the falsehood of information according to an embodiment of the present invention, such as Figure 4 As shown, the device 400 for identifying the falsehood degree of information includes:

[0139] The information acquisition module 401 is configured to receive an identification request, obtain information to be identified of a user's comments on a product according to the identification request, and convert the information to be identified into a heterogeneous graph to be identified;

[0140] The feature calculation module 402 is used to call a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified;

[0141] The identification module 403 is used to input the heterogeneous graph to be identified, the user's initial features, the comment's initial features, and the product's initial features into a trained neural network model to obtain a result of identifying the falsehood degree of the information to be identified; wherein the neural network model is trained based on historical information of user comments on products with known falsehood degrees.

[0142] The scenario in this application can be a user commenting on a product (such as goods / catering / services, etc.). In actual situations, in order to attract more users, some businesses often hire water armies to give false positive reviews to their products; at the same time, competitors may also hire water armies to give false negative reviews to specific products, thereby interfering with users' normal judgment. These false reviews can be considered abnormal reviews relative to normal user reviews. The anomaly detection algorithm is to identify these abnormal samples from normal samples. The anomaly detection method provided in this application can be applied to the above-mentioned false review detection, and can also be applied to various application scenarios such as credit card fraud, network security intrusion detection, and medical image anomaly detection.

[0143] With the rise and application of deep learning, graph neural network models are widely used in anomaly detection. Embodiments of the present invention provide a method and device for identifying the degree of information falsity. These methods convert the to-be-identified information of user product reviews into a heterogeneous graph, and then use a graph neural network model to effectively integrate the graph structure information with the characteristics of the nodes in the graph to detect and determine the degree of falsity of the reviews.

[0144] In some embodiments, before receiving the identification request, the method further includes:

[0145] Generate the identification request for all the review information in a preset time period according to a preset frequency; or

[0146] The identification request is generated based on the received identification dimension information.

[0147] The identification request in this application can be generated based on a preset frequency set in advance (such as generating a request once a week, or generating a request once every half a month, etc.), or it can be generated when an external input identification dimension is received; in some practical applications, the identification dimension may include user level, product category, time range and other information. The identification dimension can be input by the user, or it can be an identification dimension indicated in the identification instruction issued by other upstream systems; it can be flexibly set according to actual conditions.

[0148] In some embodiments, converting the information to be identified into a heterogeneous graph to be identified includes:

[0149] Determine the user node, comment node, and product node in the heterogeneous graph to be identified based on the user information, comment content, and product information in the information to be identified; wherein the comment node is the node to be identified, and the user node and the product node are adjacent nodes of the comment node;

[0150] Determine, based on the association relationship between the user, the comment, and the product indicated by the user's comment on the product in the information to be identified, the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified;

[0151] Based on the user nodes, comment nodes, product nodes, edges between user nodes and comment nodes, and edges between comment nodes and product nodes, the heterogeneous graph to be identified is constructed.

[0152] The heterogeneous graph constructed according to the above method can present the information to be identified in a specific and complete manner in the form of a heterogeneous graph without missing the association relationship between nodes, and can be easily combined with node features.

[0153] In some embodiments, the neural network model is trained according to the following method, including:

[0154] Retrieve historical information about product reviews from users with known falsehood levels, and select a preset amount of information from them as training data;

[0155] The training data is converted into a sample heterogeneous graph containing user nodes, comment nodes, and product nodes as training samples; wherein the comment node is the node to be identified, and the user node and the product node are both adjacent nodes of the comment node; and the falsehood degree of historical information in the training data is used as the sample label of the node to be identified;

[0156] Retrieving a feature calculation method set to calculate user initial features, comment initial features, and product initial features in the training data as sample features;

[0157] The neural network model is obtained by training according to the training samples, sample labels, and sample features.

[0158] In the historical information of product reviews by users with known falsehood levels, each review may include the following content information: user-product-review content-rating-review time-whether it is a false review.

[0159] In some embodiments, the training of the neural network model based on the training samples, sample labels, and sample features includes:

[0160] For each comment node in the sample heterogeneous graph, determine its adjacent user nodes and adjacent product nodes;

[0161] For the adjacent user node, determine other review nodes and other product nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial product features, call an aggregation function to generate a first user representation vector for the adjacent user node; concatenate the first user representation vector with the initial user features of the adjacent user node to obtain a user representation vector for the adjacent user node;

[0162] For the adjacent product node, determine other review nodes and other user nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial user features, and call an aggregation function to generate a first product representation vector for the adjacent product node; concatenate the first product representation vector with the initial product features of the adjacent product node to obtain a product representation vector for the adjacent product node;

[0163] The sample heterogeneous graph, user representation vector, product representation vector, initial comment features of comment nodes, and the sample labels are connected to a multi-layer neural network and trained to obtain the neural network model.

[0164] The user representation vector, product representation vector, and initial comment features of the comment node obtained by the above method can fully and detailedly express the meaning of each node without missing the association relationship between nodes or the characteristic information of the node itself, thereby making the trained model more accurate.

[0165] Furthermore, in some embodiments, calling an aggregation function to generate the first user representation vector of the adjacent user node, or calling an aggregation function to generate the first product representation vector of the adjacent product node, includes:

[0166] An attention mechanism is used in the aggregation function to generate the first user representation vector or the first product representation vector.

[0167] Using the attention mechanism enables aggregated features to more accurately describe reviews. This translates to the following: when a user reviews many products, reviews that are closer to the product currently being reviewed are more valuable for consideration. For example, if a user has reviewed sports, food, and electronics, and their current review is about food, the attention mechanism can aggregate more of the user's food reviews while ignoring irrelevant information. This allows the aggregated features to more accurately describe the reviews, leading to a more accurate trained model.

[0168] In some embodiments, for each adjacent user node of a comment node, after determining other user nodes that can be reached by the adjacent product node of the comment node in the sample heterogeneous graph, the user similarity between the adjacent user node and the other user nodes is calculated based on the initial user features, and a preset filtering method is used to determine a second user node from the other user nodes based on the user similarity, and determine a corresponding second product node; based on the second user node and the corresponding second product node, a user representation vector of the adjacent user node is determined; and / or,

[0169] For each adjacent product node of a comment node, after determining other product nodes that can be reached by the adjacent user node of the comment node in the sample heterogeneous graph, the product similarity between the adjacent product node and the other product nodes is calculated based on the initial product features, and a preset filtering method is used to determine a third product node from the other product nodes based on the product similarity, and determine a corresponding third user node; based on the third product node and the corresponding third user node, the product representation vector of the adjacent product node is determined.

[0170] The above filtering method can include setting a filtering threshold or a percentage threshold; the following examples illustrate:

[0171] When the user similarity between an adjacent user node and other user nodes is calculated to be: 1, 0.9, 0.8, 0.7, 0.2, if the filtering method is to set the filtering threshold, then when the filtering threshold is 0.6, the user nodes corresponding to 1, 0.9, 0.8, 0.7 can be taken for subsequent calculations; if the filtering method is to set the filtering percentage threshold, then when the filtering threshold is 60%, the user nodes corresponding to 1, 0.9, 0.8 can be taken for subsequent calculations; under different scenario requirements, different filtering methods can be used and different filtering parameters (such as filtering threshold or filtering percentage threshold) can be set.

[0172] Using the above method, by screening out nodes that are more similar to the node of interest to determine the representation vector, it is possible to aggregate more effective node information and express node features more accurately, thereby improving model performance.

[0173] In some embodiments, the user initial feature includes at least one of the following: a feature of the number of comments posted by the user, a feature of the level of comments posted by the user, a feature of the frequency of comments posted by the user, and a feature of the length of comments posted by the user;

[0174] The initial review feature includes at least one of the following: a grade feature, a grade deviation feature, a content feature, and a feature of the corresponding product being reviewed;

[0175] The initial product features include at least one of the following: a feature of the number of reviews, a feature of the review level, a feature of the review content, and a feature of the review frequency.

[0176] Figure 5 An exemplary system architecture 500 is shown to which the method or apparatus for identifying the falsehood level of information according to an embodiment of the present invention may be applied.

[0177] like Figure 5 As shown, system architecture 500 may include terminal devices 501, 502, 503, a network 504, and a server 505. Network 504 is used to provide a medium for communication links between terminal devices 501, 502, 503 and server 505. Network 504 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0178] Users can use terminal devices 501, 502, 503 to interact with server 505 via network 504 to receive or send messages, etc. Various client applications with the need to identify the falsehood of information can be installed on terminal devices 501, 502, 503.

[0179] The terminal devices 501 , 502 , and 503 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0180] The server 505 may be a server that provides various services, such as a background management server that supports users in identifying the falsehood of information using the terminal devices 501, 502, and 503. The background management server may analyze and process the received identification request and feed back the processing result to the terminal device.

[0181] It should be noted that the method for identifying the degree of falsehood of information provided in the embodiment of the present invention is generally executed by the server 505 , and accordingly, the device for identifying the degree of falsehood of information is generally provided in the server 505 .

[0182] It should be understood that Figure 5The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0183] Reference below Figure 6 , which shows a schematic structural diagram of a computer system 600 of a terminal device suitable for implementing an embodiment of the present invention. Figure 6 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0184] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage unit 608 into a random access memory (RAM) 603. Various programs and data required for the operation of the system 600 are also stored in the RAM 603. The CPU 601, ROM 602, and RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0185] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, and the like; an output section 607 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 608 including a hard disk; and a communication section 609 including a network interface card such as a LAN card or a modem. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 610 as needed, so that computer programs read therefrom can be installed into the storage section 608 as needed.

[0186] In particular, according to the embodiments disclosed in the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 609, and / or installed from a removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the system of the present invention are performed.

[0187] It should be noted that the computer-readable medium described in the present invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media can include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In the present invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. This propagated data signal can take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wireline, optical fiber cable, RF, or any suitable combination thereof.

[0188] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0189] The modules described in the embodiments of the present invention may be implemented in software or hardware. The modules described may also be provided in a processor. For example, a processor may be described as including an information acquisition module, a feature calculation module, and an identification module. The names of these modules do not, in some cases, limit the modules themselves.

[0190] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently and not be assembled into the device. The above computer-readable medium carries one or more programs. When the above one or more programs are executed by a device, the device includes: step S101, receiving an identification request, obtaining information to be identified of a user's comments on a product according to the identification request, and converting the information to be identified into a heterogeneous graph to be identified; step S102, calling a feature calculation method set, calculating the user initial features, comment initial features, and product initial features in the information to be identified; step S103, inputting the heterogeneous graph to be identified, the user initial features, the comment initial features, and the product initial features into a trained neural network model to obtain a false degree identification result of the information to be identified; wherein, the neural network model is trained based on historical information of user comments on products with known false degrees.

[0191] According to the technical solution of the embodiment of the present invention, because the information to be identified of user comments on products is converted into a heterogeneous graph, and then a graph neural network model is used to effectively fuse the graph structure information and the characteristics of the nodes in the graph for detection to determine the degree of falsehood of the comments, it overcomes the technical problem that the graph neural network models in the prior art are mostly applicable to homogeneous graphs or do not consider the characteristics of the graph nodes themselves, and are therefore not applicable to the comment detection scenario in the present invention, thereby achieving the technical effect of being able to use the graph neural network model combined with the characteristics of the comment information itself to detect false comments.

[0192] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for identifying the degree of false information, characterized in that: include: Receive an identification request, obtain information to be identified of a user's comments on a product according to the identification request, and convert the information to be identified into a heterogeneous graph to be identified; wherein converting the information to be identified into a heterogeneous graph to be identified includes: determining the user node, comment node, and product node in the heterogeneous graph to be identified respectively according to each user information, comment content, and product information in the information to be identified; wherein the comment node is a node to be identified, and the user node and the product node are adjacent nodes of the comment node; according to the association relationship between the user, comment, and product indicated by the user's comments on the product in the information to be identified, determine the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified; based on the user node, comment node, product node, the edge between the user node and the comment node, and the edge between the comment node and the product node, construct the heterogeneous graph to be identified shown. Retrieving a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified; The heterogeneous graph to be identified, the initial features of users, the initial features of comments, and the initial features of products are input into a trained neural network model to obtain a result of identifying the degree of falsity of the information to be identified; wherein the neural network model is obtained by training a multi-layer neural network based on historical information of users' comments on products with known degrees of falsity; for a product node, other comment nodes and other user nodes that the node can reach are determined in the heterogeneous graph to be identified, and the initial features of comments and the initial features of users are combined to form conversion data for splicing with the initial features of the product to obtain a product representation vector.

2. The method according to claim 1, characterized in that Before receiving the identification request, it also includes: Generate the identification request for all the review information in a preset time period according to a preset frequency; or The identification request is generated based on the received identification dimension information.

3. The method according to claim 1, characterized in that The neural network model is trained according to the following method, including: Retrieve historical information about product reviews from users with known falsehood levels, and select a preset amount of information from them as training data; The training data is converted into a sample heterogeneous graph containing user nodes, comment nodes, and product nodes as training samples; wherein the comment node is the node to be identified, and the user node and the product node are both adjacent nodes of the comment node; and the falsehood degree of historical information in the training data is used as the sample label of the node to be identified; Retrieving a feature calculation method set to calculate user initial features, comment initial features, and product initial features in the training data as sample features; The neural network model is obtained by training according to the training samples, sample labels, and sample features.

4. The method according to claim 3, characterized in that The neural network model is obtained by training according to the training samples, sample labels, and sample features, including: For each comment node in the sample heterogeneous graph, determine its adjacent user nodes and adjacent product nodes; For the adjacent user node, determine other review nodes and other product nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial product features, call an aggregation function to generate a first user representation vector for the adjacent user node; concatenate the first user representation vector with the initial user features of the adjacent user node to obtain a user representation vector for the adjacent user node; For the adjacent product node, determine other review nodes and other user nodes that the node can reach in the sample heterogeneous graph, combine the initial review features and the initial user features, and call an aggregation function to generate a first product representation vector for the adjacent product node; concatenate the first product representation vector with the initial product features of the adjacent product node to obtain a product representation vector for the adjacent product node; The sample heterogeneous graph, user representation vector, product representation vector, initial comment features of comment nodes, and the sample labels are connected to a multi-layer neural network and trained to obtain the neural network model.

5. The method according to claim 4, characterized in that Calling an aggregation function to generate a first user representation vector of the adjacent user node, or calling an aggregation function to generate a first product representation vector of the adjacent product node, includes: An attention mechanism is used in the aggregation function to generate the first user representation vector or the first product representation vector.

6. The method according to claim 4, characterized in that Also includes: For each comment node's adjacent user node, after determining other user nodes that can be reached by the comment node's adjacent product node in the sample heterogeneous graph, calculate the user similarity between the adjacent user node and the other user nodes based on the user's initial features, and use a preset filtering method to determine a second user node from the other user nodes based on the user similarity, and determine a corresponding second product node; Determine a user representation vector of the adjacent user node according to the second user node and the corresponding second product node; and / or, For each adjacent product node of a comment node, after determining other product nodes that can be reached by the adjacent user node of the comment node in the sample heterogeneous graph, the product similarity between the adjacent product node and the other product nodes is calculated based on the initial product features, and a preset filtering method is used to determine a third product node from the other product nodes based on the product similarity, and determine a corresponding third user node; based on the third product node and the corresponding third user node, the product representation vector of the adjacent product node is determined.

7. The method according to claim 1 or 3, characterized in that The user initial feature includes at least one of the following: a feature of the number of comments posted by the user, a feature of the level of comments posted by the user, a feature of the frequency of comments posted by the user, and a feature of the length of comments posted by the user; The initial review feature includes at least one of the following: a grade feature, a grade deviation feature, a content feature, and a feature of the corresponding product being reviewed; The initial product features include at least one of the following: a feature of the number of reviews, a feature of the review level, a feature of the review content, and a feature of the review frequency.

8. A device for identifying the degree of false information, characterized in that: include: An information acquisition module is used to receive an identification request, obtain information to be identified of a user's comments on a product according to the identification request, and convert the information to be identified into a heterogeneous graph to be identified; wherein, converting the information to be identified into a heterogeneous graph to be identified includes: determining the user node, comment node, and product node in the heterogeneous graph to be identified according to each user information, comment content, and product information in the information to be identified; wherein the comment node is the node to be identified, and the user node and the product node are adjacent nodes of the comment node; according to the association relationship between the user, comment, and product indicated by the user's comments on the product in the information to be identified, determining the edge between the user node and the comment node, and the edge between the comment node and the product node in the heterogeneous graph to be identified; based on the user node, comment node, product node, the edge between the user node and the comment node, and the edge between the comment node and the product node, constructing the heterogeneous graph to be identified shown. A feature calculation module is used to call a feature calculation method set to calculate the user initial features, comment initial features, and product initial features in the information to be identified; The identification module is used to input the heterogeneous graph to be identified, the initial features of users, the initial features of comments, and the initial features of products into a trained neural network model to obtain an identification result of the degree of falsity of the information to be identified; wherein the neural network model is obtained by training a multi-layer neural network based on historical information of users' comments on products with known degrees of falsity; for a product node, other comment nodes and other user nodes that the node can reach are determined in the heterogeneous graph to be identified, and the initial features of comments and the initial features of users are combined to form conversion data for splicing with the initial features of the product to obtain a product representation vector.

9. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images