Method for detecting authenticity of network public opinion of platform enterprise based on time sequence knowledge graph

Abstract

The application discloses a platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph. The platform enterprise network public opinion authenticity detection method comprises the following steps: splitting a platform enterprise time sequence knowledge graph into multiple static knowledge graphs connected in series in a time dimension; performing vectorization processing on the multiple static knowledge graphs; obtaining a to-be-detected public opinion text, and constructing a to-be-detected time sequence knowledge graph; splitting the to-be-detected time sequence knowledge graph into at least one to-be-detected static knowledge graph connected in series in the time dimension; performing vectorization processing on the at least one to-be-detected static knowledge graph, and constructing a to-be-detected public opinion time sequence knowledge graph vector; obtaining a subgraph similar in semantics to the to-be-detected public opinion time sequence knowledge graph vector from the platform enterprise time sequence knowledge graph vector, and determining the authenticity of the to-be-detected public opinion text according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector. The application improves the accuracy of the determination and effectively reduces the calculation cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of network public opinion detection, and in particular to a platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph. BACKGROUND

[0002] With the continuous development and application of digital technology, traditional industry models and business models are being overturned and restructured. Digital economy has become an important force driving economic development and social progress. As an important part of digital economy, platform enterprises promote the production, circulation and consumption of digital products and services through the establishment and operation of various digital platforms, and promote the rapid development and popularization of digital economy.

[0003] With the rise of social media, the Internet has become an important channel for people to access information, express emotions and exchange ideas. However, with the increasing speed of information dissemination, the authenticity and accuracy of network public opinion have been questioned more and more. Especially for platform enterprises, due to the particularity of their influence and discourse power, the generation, expansion and spread of platform enterprise public opinion will have an important impact on investors, platform enterprises, market supervision and even macroeconomic operation. Therefore, the detection of network public opinion authenticity is particularly important: on the one hand, it can protect the reputation and image of platform enterprises and avoid damage to the image of platform enterprises by false information, and on the other hand, it can maintain the right to know of the public and ensure the authenticity and objectivity of network public opinion, thereby promoting the stable and healthy development of digital economy.

[0004] Platform enterprise public opinion information is generally in the form of text. At present, the methods for detecting the authenticity of public opinion text are one based on rules, which uses artificially designed rules or patterns to detect fake news. This method can quickly determine some fake news, but it cannot detect new types of fake news that do not conform to known rules, and the generalization effect is poor. The other is based on machine learning, which trains known true and false news samples to build a classification model to classify news as true or false. This method can adaptively detect new fake news, but it requires a large amount of training data, has high requirements for data quality, and the model may have problems such as overfitting or underfitting.

[0005] The above methods only focus on the content of public opinion text, which usually contains a large number of entities. These entities and the relationships between them play a big role in determining authenticity, but these entities and relationships cannot be effectively understood by the model directly from the public opinion text, and there may be ambiguity, and the interpretability of the determination result is poor. SUMMARY

[0006] The embodiment of the present application provides a platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph, to solve the problem of inaccurate detection in the prior art based on rules and machine learning.

[0007] The platform enterprise network public opinion authenticity detection method based on the time sequence knowledge graph comprises:

[0008] The platform enterprise time sequence knowledge graph is acquired, and the platform enterprise time sequence knowledge graph is split into a plurality of static knowledge graphs connected in series in the time dimension, the static knowledge graph is in a quadruple form, and the quadruple form comprises a head entity, a tail entity, a relationship and a timestamp;

[0009] The plurality of static knowledge graphs are subjected to vectorization processing by using a graph embedding module method, so as to construct a platform enterprise time sequence knowledge graph vector;

[0010] A to-be-detected public opinion text is acquired, and a to-be-detected time sequence knowledge graph is constructed based on the to-be-detected public opinion text;

[0011] The to-be-detected time sequence knowledge graph is split into at least one to-be-detected static knowledge graph connected in series in the time dimension, and the to-be-detected static knowledge graph is in a quadruple form;

[0012] The at least one to-be-detected static knowledge graph is subjected to vectorization processing by using a graph embedding module method, so as to construct a to-be-detected public opinion time sequence knowledge graph vector;

[0013] A subgraph similar in semantics to the to-be-detected public opinion time sequence knowledge graph vector is acquired from the platform enterprise time sequence knowledge graph vector by using a cosine similarity of semantic representation, and the authenticity of the to-be-detected public opinion text is determined according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0014] Corresponding to the above method, the embodiment of the present application further provides a platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph, comprising:

[0015] A platform enterprise time sequence knowledge graph vector representation module is configured to acquire a platform enterprise time sequence knowledge graph, and split the platform enterprise time sequence knowledge graph into a plurality of static knowledge graphs connected in series in the time dimension, the static knowledge graph is in a quadruple form, and the quadruple form comprises a head entity, a tail entity, a relationship and a timestamp; the plurality of static knowledge graphs are subjected to vectorization processing by using a graph embedding module method, so as to construct a platform enterprise time sequence knowledge graph vector;

[0016] The to-be-detected public opinion time sequence knowledge graph construction module is configured to obtain a to-be-detected public opinion text and construct a to-be-detected time sequence knowledge graph based on the to-be-detected public opinion text; and the to-be-detected time sequence knowledge graph is split into at least one to-be-detected static knowledge graph which is in series in the time dimension, and the to-be-detected static knowledge graph is in the form of a quadruple;

[0017] The to-be-detected public opinion time sequence knowledge graph vector representation module is configured to perform vectorization processing on the at least one to-be-detected static knowledge graph by using a graph embedding module method, so as to construct a to-be-detected public opinion time sequence knowledge graph vector.

[0018] The graph similarity detection module is configured to obtain a subgraph which is semantically similar to the to-be-detected public opinion time sequence knowledge graph vector from the platform enterprise time sequence knowledge graph vector by using a cosine similarity of semantic representation, and determine a similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0019] The public opinion authenticity joint detection module is configured to determine the authenticity of the to-be-detected public opinion text according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0020] The embodiment of the present application also provides a platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph, which comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the computer program implements the steps of the platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph when executed by the processor.

[0021] The embodiment of the present application also provides a computer readable storage medium, which stores an information transmission implementation program, and the program implements the steps of the platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph when executed by a processor.

[0022] By fully utilizing time information to assist in public opinion authenticity discrimination, the embodiment of the present application improves the accuracy of discrimination and effectively reduces the computational cost.

[0023] The above description is only a summary of the technical scheme of the present application, in order to more clearly understand the technical means of the present application, the specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0024] Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present application. In the drawings:

[0025] Figure 1 FIG. 1 is a schematic diagram of a platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph according to an embodiment of the present application. DETAILED DESCRIPTION

[0026] Exemplary embodiments of the present application will be described herein below with reference to the accompanying drawings. While exemplary embodiments of the present application are shown in the drawings, it is understood that the present application can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that this application will be thoroughly and completely understood, and so that the scope of the present application will be completely conveyed to those skilled in the art. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure the understanding of this description.

[0027] A knowledge graph is a new knowledge representation paradigm that stores knowledge using nodes and relationships, and is usually expressed in the form of triples (entity, relationship, entity). The knowledge graph has great advantages in the semantic, logical and relational nature of knowledge, and various knowledge graphs have emerged in vertical fields. In recent years, with the rapid development of knowledge graph technology, the introduction of knowledge graphs as external knowledge for public opinion text authenticity detection has also gradually emerged.

[0028] Public opinion authenticity detection based on a knowledge graph can effectively improve the discrimination effect and enhance the interpretability of the results. However, public opinion of platform enterprises in the real world has strong time characteristics, which are reflected in two aspects: the time of public opinion generation and the time mentioned in the public opinion text content. Time information is of great significance in public opinion text authenticity detection. However, a knowledge graph is static, and the entities and relationships in the knowledge graph may only be valid for a period of time, such as the triple (Tencent, product, Tencent Weibo), which is only valid from April 1, 2010 to September 28, 2020. When determining the authenticity of a related public opinion text, for example, "Xinhua News Agency, March 25, 2021, this morning, Tencent QQ announced through Tencent Weibo that there will be a large-scale QQ password theft by unscrupulous individuals in the near future, please change your password in time", if the enterprise knowledge graph triple (Tencent, product, Tencent Weibo) is used, the authenticity cannot be determined.

[0029] Based on this, the present application proposes a platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph, as well as a corresponding device, equipment and computer readable storage medium.

[0030] The time sequence knowledge graph is usually represented by a quadruple (entity, relationship, entity, timestamp). Compared with the static knowledge graph, the time sequence knowledge graph has a time validity period of entity and relationship, so that the knowledge in the knowledge graph is more realistic. These time information can assist detection in combination with the time characteristics of public opinion text, effectively improving the effect of public opinion text authenticity detection. For example, for the above public opinion text, based on the enterprise time sequence knowledge graph information: (Tencent, product, Tencent microblog, [2010-4-1, 2020-9-28]), that is, the effective time of Tencent microblog is from April 1, 2010 to September 28, 2020, and the time mentioned in the public opinion text is March 25, 2021, so it can be determined that the public opinion information is false.

[0031] The platform enterprise network public opinion authenticity detection method based on the time sequence knowledge graph of the embodiment of the application comprises:

[0032] The platform enterprise time sequence knowledge graph is obtained. If the platform enterprise maintains the platform enterprise time sequence knowledge graph, the platform enterprise time sequence knowledge graph can be directly obtained, and if not, the platform enterprise time sequence knowledge graph can be constructed by itself.

[0033] The platform enterprise time sequence knowledge graph is split into a plurality of static knowledge graphs connected in the time dimension. The static knowledge graph is in the form of a quadruple, and the quadruple form includes a head entity, a tail entity, a relationship and a timestamp.

[0034] The plurality of static knowledge graphs are vectorized by using a graph embedding module method to construct a platform enterprise time sequence knowledge graph vector.

[0035] The public opinion text to be detected is obtained, and a to-be-detected time sequence knowledge graph is constructed based on the public opinion text to be detected.

[0036] The to-be-detected time sequence knowledge graph is split into at least one to-be-detected static knowledge graph connected in the time dimension. The to-be-detected static knowledge graph is in the form of a quadruple.

[0037] It should be noted that if the to-be-detected time sequence knowledge graph itself is a time-dimension static knowledge graph, it does not need to be split, and the constructed time sequence knowledge graph is a to-be-detected static knowledge graph. If the to-be-detected time sequence knowledge graph constructed is continuous in the time dimension, the splitting step needs to be performed. The specific operation process is performed according to the actual situation.

[0038] The at least one to-be-detected static knowledge graph is vectorized by using a graph embedding module method to construct a to-be-detected public opinion time sequence knowledge graph vector.

[0039] The subgraph is obtained from the platform enterprise time sequence knowledge graph vector according to the cosine similarity of the semantic representation, and the authenticity of the to-be-detected public opinion text is determined according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0040] The platform enterprise time sequence knowledge graph vector maintains true public opinion information, the subgraph obtained from the platform enterprise time sequence knowledge graph vector is most relevant to the semantic representation of the to-be-detected public opinion time sequence knowledge graph vector, if the to-be-detected public opinion time sequence knowledge graph vector has high similarity with the subgraph, it is indicated that the authenticity is high, if the similarity is low, it is indicated that the representation is basically opposite to the truth, and then it is indicated that it is false.

[0041] Therefore, by means of the time sequence knowledge graph, the platform enterprise network public opinion authenticity machine can be detected, the detection process fully utilizes time information to assist in public opinion authenticity discrimination, improves the accuracy of discrimination, and compared with the detection method in the prior art, the calculation process is simple, and the calculation overhead is effectively reduced.

[0042] On the basis of the above embodiment, further variant embodiments are proposed, and it should be noted that, in order to make the description brief, only the differences from the above embodiment are described in the variant embodiments.

[0043] According to some embodiments of the application, the method further comprises:

[0044] The to-be-detected time sequence knowledge graph is subjected to time sequence consistency detection based on at least one constraint condition in a time non-overlapping constraint, a time sequence constraint and a time validity period constraint.

[0045] The time non-overlapping constraint: for a relationship with unique properties, the time interval of any two facts with the same relationship and the same head entity is not overlapped, for example, the “legal representative” of a company can only be one person at the same time.

[0046] The time sequence constraint: for some relationships with time attributes, one fact always occurs before another fact, for example, a company must be “registered” in a certain place before it can be “listed” on a certain stock exchange. The time sequence constraint ensures that the order of occurrence of certain relationship pairs conforms to the reality logic.

[0047] The time validity period constraint: some facts are only true within a certain time period, for example, “Tencent's product Tencent Weibo” is only valid from April 1, 2010 to September 28, 2020. The time validity period constraint ensures that some facts are true within a certain time period.

[0048] Based on the time sequence consistency detection result, the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector is combined to comprehensively determine the authenticity of the to-be-detected public opinion text.

[0049] Therefore, the time sequence consistency constraint can be applied to guide the true and false public opinion discrimination.

[0050] Further, based on the time sequence consistency detection result, the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector is combined to comprehensively determine the authenticity of the to-be-detected public opinion text.

[0051] The similarity score of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector is calculated, and it is determined whether the similarity score is lower than a threshold value. If yes, it is determined that the to-be-detected public opinion text is false, otherwise, the authenticity of the to-be-detected public opinion text is further determined based on the time sequence consistency detection result.

[0052] It should be noted that if the authenticity of the public opinion text has been determined through the similarity score, the time sequence consistency detection is not needed, and the time sequence consistency detection is an auxiliary means.

[0053] According to some embodiments of the present application, the plurality of static knowledge graphs are vectorized to construct a platform enterprise time sequence knowledge graph vector.

[0054] A graph convolution network (GCN) is learned for each static knowledge graph to generate an embedding representation of the static knowledge graph.

[0055] A GRU model is used to learn the static knowledge graphs in series to form a platform enterprise time sequence knowledge graph vector.

[0056] In some embodiments of the present application, the construction of the time sequence knowledge graph can use the named entity recognition function in the hanlp tool to carry out named entity recognition, and use the dependency syntax analysis in the hanlp tool to carry out relationship extraction and time attribute extraction, and finally generate a four-tuple set with time annotation.

[0057] It should be noted that the above-provided method is not inherently related to any specific computer, virtual system or other device. Various general-purpose systems can also be used with the teaching based on the above. According to the above description, the structure required to construct such a system is obvious. In addition, the present application is not directed to any specific programming language. It should be understood that the content of the present application described herein can be implemented using various programming languages, and the above description of a specific language is to disclose the best embodiment of the present application.

[0058] Corresponding to the above method, the embodiment of the application also proposes a platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph, comprising:

[0059] A platform enterprise time sequence knowledge graph vector representation module is configured to obtain a platform enterprise time sequence knowledge graph, and split the platform enterprise time sequence knowledge graph into a plurality of static knowledge graphs connected in series in a time dimension, wherein the static knowledge graphs are in a quadruple form, and the quadruple form includes a head entity, a tail entity, a relationship, and a timestamp; a graph embedding module method is used to perform vectorization processing on the plurality of static knowledge graphs to construct a platform enterprise time sequence knowledge graph vector.

[0060] A to-be-detected public opinion time sequence knowledge graph construction module is configured to obtain a to-be-detected public opinion text, and construct a to-be-detected time sequence knowledge graph based on the to-be-detected public opinion text; the to-be-detected time sequence knowledge graph is split into at least one to-be-detected static knowledge graph connected in series in a time dimension, wherein the to-be-detected static knowledge graph is in a quadruple form.

[0061] A to-be-detected public opinion time sequence knowledge graph vector representation module is configured to use a graph embedding module method to perform vectorization processing on the at least one to-be-detected static knowledge graph to construct a to-be-detected public opinion time sequence knowledge graph vector.

[0062] A graph similarity detection module is configured to use a cosine similarity of semantic representation to obtain a subgraph similar in semantics to the to-be-detected public opinion time sequence knowledge graph vector from the platform enterprise time sequence knowledge graph vector, and determine a similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0063] A public opinion authenticity joint detection module is configured to determine authenticity of the to-be-detected public opinion text according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0064] According to some embodiments of the application, the device further comprises:

[0065] A time sequence consistency detection module is configured to perform time sequence consistency detection on the to-be-detected time sequence knowledge graph based on at least one constraint condition of a time non-overlapping constraint, a time sequence constraint, and a time validity period constraint.

[0066] The public opinion authenticity joint detection module is configured to determine authenticity of the to-be-detected public opinion text based on a time sequence consistency detection result and in combination with the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

[0067] Further, the public opinion authenticity joint detection module is configured to:

[0068] The similarity score of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector is calculated, whether the similarity score is lower than a threshold is judged, if yes, it is determined that the to-be-detected public opinion text is false, otherwise, the authenticity of the to-be-detected public opinion text is further determined based on the time sequence consistency detection result.

[0069] According to some embodiments of the application, the platform enterprise time sequence knowledge graph vector representation module is used to:

[0070] A graph convolution network (GCN) is learned for each static knowledge graph to generate an embedding representation of the static knowledge graph.

[0071] A GRU model is used to learn the static knowledge graphs in series to form a platform enterprise time sequence knowledge graph vector.

[0072] A platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph according to embodiments of the application is described in detail below with reference to the drawings in one specific embodiment. It should be understood that the following description is only exemplary and should not be construed as a specific limitation of the application.

[0073] The platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph according to embodiments of the application realizes the function of performing semantic analysis on platform enterprise public opinion text s to be detected for authenticity, and constructing a to-be-detected public opinion time sequence knowledge graph Based on a given platform enterprise time sequence knowledge graph The similarity between and is compared, and a time sequence consistency constraint test is performed, so as to determine the authenticity of the public opinion text s.

[0074] Referring to Figure 1 , the platform enterprise network public opinion authenticity detection device based on a time sequence knowledge graph according to embodiments of the application includes:

[0075] (1) a platform enterprise time sequence knowledge graph vector representation module based on a graph neural network;

[0076] The time sequence knowledge graph extends the traditional static knowledge graph in the time dimension. According to embodiments of the application, the platform enterprise time sequence knowledge graph is divided into multiple static knowledge graphs in the time dimension based on the time information on the relationship edges between nodes, that is, the time sequence knowledge graph is equivalent to a series connection of multiple static knowledge graphs in the time dimension.

[0077] A given enterprise time sequence knowledge graph is represented by a four-tuple (e i , r, e j , t) with time information: e i and e jhas a relation r during the time interval t = [t b , t e ] where t b < t e . For some facts that occur at a certain time but do not last, set t b = t e . For some facts that have not ended, express t as t = [t b , +∞].

[0078] First, learn the graph convolution network GCN for each time slice of the static knowledge graph separately. Since the adjacency matrix of each static knowledge graph is different, the input of each GCN is also different, and the embedding representation of the static knowledge graph is obtained. Subsequently, the GRU model is used to learn the parameters of the static knowledge graph in series. The node representation of the current time is calculated by the embedding representation of the node and the parameter of the last time . The embedding representation update formula of the time sequence graph is:

[0079]

[0080] This calculation formula is the common calculation method of the GRU model.

[0081] Finally, the platform enterprise time sequence knowledge graph vector representation

[0082] (2) The to-be-detected public opinion time sequence knowledge graph construction module;

[0083] For the to-be-detected public opinion text s, construct the time sequence knowledge graph, use the named entity recognition function in the hanlp tool to carry out named entity recognition, and use the dependency syntax analysis in the hanlp tool to carry out relation extraction and time attribute extraction work, and finally generate a set of four-tuple with time annotation (e i , r, e j , t), that is, the to-be-detected public opinion time sequence knowledge graph

[0084] (3) The to-be-detected public opinion time sequence knowledge graph vector representation module;

[0085] Based on the graph embedding module method, the to-be-detected public opinion time sequence knowledge graph is divided into multiple static knowledge graphs in the time dimension, embedded by the graph convolution network GCN model, and the multiple static knowledge graphs are connected in series in the time dimension by the GRU, and finally the to-be-detected public opinion time sequence knowledge graph vector representation

[0086] (4) The graph similarity detection module;

[0087] Cosine similarity of semantic representation is used to find the input time knowledge graph to be detected public opinion Subgraphs with similar semantics The learned subgraph and the embedding representation of the time knowledge graph to be detected public opinion are then calculated for similarity, and the authenticity detection is completed.

[0088]

[0089] The similarity calculation score is finally obtained, and the result is a floating point number before the interval (0, 1].

[0090] (5) Time consistency detection module;

[0091] Although the embedding representation used in the similarity calculation process has captured the inherent time information, it still does not have exact time information constraints, and subtle time errors are not obvious in similarity calculation, which may lead to detection errors. By applying time consistency constraints, exact "0" and "1" judgments are given to guide the authenticity of public opinion discrimination. The present application considers three kinds of time sequence constraints: time non-overlapping, time sequence and time validity period.

[0092] Time non-overlapping constraint: for relationships with unique properties, the time interval of any two facts with the same relationship and the same head entity is non-overlapping, such as a company's "legal representative" can only be one person at the same time. Time non-overlapping constraint avoids the divergence between two facts with the same head entity and relationship, expressed as:

[0093]

[0094] Where is a Boolean decision variable to indicate whether fact (e i ,r k ,e j ,t) is true for time t. is a set of relationships described. Here has the same meaning as , but the subscript is different, indicating a Boolean decision variable to indicate whether fact (e i ,r k ,e l ,t) is true for time t.

[0095] Time sequence constraint: for some relationships with time attributes, one fact always occurs before another fact, such as a company must be "registered" in a certain place before it can be "listed" on a certain exchange. Time sequence constraint ensures that the order of appearance of certain relationship pairs conforms to the reality logic, which can be expressed as: ​

[0096]

[0097] wherein {<r k ,r' k >} is a pair of relations with sequential order, such as <registered address, stock exchange>. These pairs of relations are discovered and marked from the platform enterprise time knowledge graph by statistical and artificial methods.

[0098] Time validity period constraint: some facts are only true in a certain time period, such as "Tencent's product Tencent Weibo" is only valid from 2010-4-1 to 2020-9-28. The time validity period constraint ensures that some facts are true in a certain time period, and these constraints can be expressed as:

[0099]

[0100] wherein is a relation valid only in a certain time period, t Δ is the time validity period of the fact.

[0101] Joint three constraints to build a time graph for the detected public opinion text Consistency detection, and finally obtain The result is an integer "0" or "1".

[0102] (6) Public opinion true and false joint detection module

[0103] The final result of the public opinion true and false detection performed by the present application is "false", not "true". The joint detection module performs true and false joint detection based on the similarity calculation score and the time consistency detection result.

[0104] When the similarity score is lower than the given threshold, it means that the detected public opinion text and the existing knowledge reserve are not similar, and it is judged as "false"; when it is higher than the threshold, it means that the detected public opinion text and the existing knowledge reserve are similar, but it is not judged as "true", and further time consistency detection is required. The final detection result gives a similar subgraph for comparison, which assists manual judgment and also improves the interpretability of the result.

[0105] When the time consistency detection result is "0", it means that the time information does not meet the consistency constraint, and it is judged as "false"; when it is "1", it means that the time information meets the consistency constraint, but it is not judged as "true". The final detection result gives detailed constraint detection results, which assist manual judgment and also improve the interpretability of the result.

[0106] The embodiment of the present application embeds all elements (including entities, relations and time, etc.) into a unified semantic space, realizes the measurement of the similarity between the detected text and the existing knowledge base in the same semantic space.

[0107] The time sequence graph embedding method provided by the embodiment of the application decomposes a time sequence graph into a series of static knowledge graphs for embedding, can capture a nonlinear time sequence evolution process, and obtain richer semantics.

[0108] The three time sequence information consistency constraints can effectively focus on true and false discrimination information from the time dimension, reduce the range of public opinion true and false judgment information acquisition, and have the advantage of reducing the calculation cost.

[0109] By performing semantic analysis on the to-be-inspected public opinion text, constructing a to-be-inspected public opinion time sequence knowledge graph, and comparing the two time sequence graphs based on the given platform enterprise time sequence graph, and according to the time sequence consistency constraint inspection result in the to-be-inspected time sequence knowledge graph, the true and false of the public opinion text is jointly judged, the time information is fully utilized to assist the public opinion true and false discrimination, the phenomenon that the time information is ignored when the external knowledge base is used to judge the public opinion true and false in the past is eliminated, the discrimination accuracy is improved, and the calculation cost is effectively reduced.

[0110] It should be noted that the above only describes the preferred embodiments of the application and is not intended to limit the application. For those skilled in the art, the application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application shall be included in the protection scope of the application.

[0111] The embodiment of the application also provides a platform enterprise network public opinion true and false detection device based on a time sequence knowledge graph, which comprises a memory, a processor and a computer program stored on the memory and executable on the processor, and the computer program implements the steps of the platform enterprise network public opinion true and false detection method based on the time sequence knowledge graph when executed by the processor.

[0112] The embodiment of the application also provides a computer readable storage medium, and the computer readable storage medium stores an information transmission implementation program, and the program implements the steps of the platform enterprise network public opinion true and false detection method based on the time sequence knowledge graph when executed by a processor.

[0113] Computer-readable storage media can take the form of, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of computer-readable storage media can include, but are not limited to, the following: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0114] Computer-readable storage media can be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a propagated data signal with computer-readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated signal can take any of a variety of forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium that is not a computer-readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable signal medium can be transmitted using any appropriate medium, including but not limited to wireless, wired, optical fiber cable, R.F, or the like, or any suitable combination of the foregoing.

[0115] It should be noted that any of the methods described herein can include an additional step of modifying the sequence of the method to conform with the language of the claims, including adding or removing steps.

[0116] The terms "comprises", "comprising", or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0117] No reference sign is to be construed as a limitation on the claim except for those signs following a chemical formula. The use of the terms "first", "second" and other such terms does not imply any order other than the order of description. Such terms can be understood as names.

Claims

1. A method for detecting the authenticity of network public opinion of a platform enterprise based on a time-series knowledge graph, characterized in that, The method comprises the following steps: acquire a platform enterprise time sequence knowledge graph, and split the platform enterprise time sequence knowledge graph into a plurality of static knowledge graphs connected in series in a time dimension, the static knowledge graphs being in a quadruple form, the quadruple form comprising a head entity, a tail entity, a relationship, and a timestamp; perform vectorization processing on the plurality of static knowledge graphs by using a graph embedding module method to construct a platform enterprise time sequence knowledge graph vector; acquire a to-be-detected public opinion text, and construct a to-be-detected time sequence knowledge graph based on the to-be-detected public opinion text; split the to-be-detected time sequence knowledge graph into at least one to-be-detected static knowledge graph connected in series in a time dimension, the to-be-detected static knowledge graph being in a quadruple form; perform vectorization processing on the at least one to-be-detected static knowledge graph by using a graph embedding module method to construct a to-be-detected public opinion time sequence knowledge graph vector; acquire a subgraph semantically similar to the to-be-detected public opinion time sequence knowledge graph vector from the platform enterprise time sequence knowledge graph vector by using a cosine similarity of a semantic representation, and determine the authenticity of the to-be-detected public opinion text according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector. The method further comprises: perform time sequence consistency detection on the to-be-detected time sequence knowledge graph based on at least one constraint condition in a time non-overlapping constraint, a time sequence constraint, and a time validity period constraint; determine the authenticity of the to-be-detected public opinion text based on the time sequence consistency detection result and the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

2. The method of claim 1, wherein, The determination of the authenticity of the to-be-detected public opinion text based on the time sequence consistency detection result and the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector comprises: calculate a similarity score of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector, determine whether the similarity score is lower than a threshold value, if yes, determine that the to-be-detected public opinion text is false, and if not, further determine the authenticity of the to-be-detected public opinion text based on the time sequence consistency detection result.

3. The method of claim 1, wherein, The vectorization processing on the plurality of static knowledge graphs by using the graph embedding module method to construct the platform enterprise time sequence knowledge graph vector comprises: individually learn a graph convolution network (GCN) for each static knowledge graph to generate an embedding representation of the static knowledge graph; learn the static knowledge graphs in series by using a GRU model to form a platform enterprise time sequence knowledge graph vector.

4. A platform enterprise network public opinion true and false detection device based on a timing knowledge graph, characterized by, The method comprises the following steps: a platform enterprise time sequence knowledge graph vector representation module is configured to acquire a platform enterprise time sequence knowledge graph, and split the platform enterprise time sequence knowledge graph into a plurality of static knowledge graphs connected in series in a time dimension, the static knowledge graphs being in a quadruple form, the quadruple form comprising a head entity, a tail entity, a relationship, and a timestamp; and perform vectorization processing on the plurality of static knowledge graphs by using a graph embedding module method to construct a platform enterprise time sequence knowledge graph vector; The to-be-detected public opinion time sequence knowledge graph construction module is configured to obtain a to-be-detected public opinion text and construct a to-be-detected time sequence knowledge graph based on the to-be-detected public opinion text; and the to-be-detected time sequence knowledge graph is split into at least one to-be-detected static knowledge graph which is in series in the time dimension, and the to-be-detected static knowledge graph is in the form of a quadruple; The to-be-detected public opinion time sequence knowledge graph vector representation module is configured to perform vectorization processing on the at least one to-be-detected static knowledge graph by using a graph embedding module method, so as to construct a to-be-detected public opinion time sequence knowledge graph vector. The graph similarity detection module is configured to obtain a subgraph that is semantically similar to the to-be-detected public opinion time sequence knowledge graph vector from the platform enterprise time sequence knowledge graph vector by using a cosine similarity of semantic representation, and determine a similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector. The public opinion authenticity joint detection module is configured to determine the authenticity of the to-be-detected public opinion text according to the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector. The device further comprises: The time sequence consistency detection module is configured to perform time sequence consistency detection on the to-be-detected time sequence knowledge graph based on at least one constraint condition of a time non-overlapping constraint, a time sequence constraint, and a time validity period constraint. The public opinion authenticity joint detection module is configured to determine the authenticity of the to-be-detected public opinion text based on the time sequence consistency detection result and the similarity degree of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector.

5. The apparatus of claim 4, wherein, The public opinion authenticity joint detection module is configured to: Calculate a similarity score of the subgraph and the to-be-detected public opinion time sequence knowledge graph vector, determine whether the similarity score is lower than a threshold value, if yes, determine that the to-be-detected public opinion text is false, and if not, further determine the authenticity of the to-be-detected public opinion text based on the time sequence consistency detection result.

6. The apparatus of claim 4, wherein, The platform enterprise time sequence knowledge graph vector representation module is configured to: Learn a graph convolutional network (GCN) for each static knowledge graph separately to generate an embedding representation of the static knowledge graph; Learn the static knowledge graphs in series by using a GRU model to form a platform enterprise time sequence knowledge graph vector.

7. A platform enterprise network public opinion true and false detection equipment based on a timing knowledge graph, characterized by, The device further comprises: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the computer program is executed by the processor to implement the steps of the platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph according to any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores an implementation program of information transmission, and the program is executed by the processor to implement the steps of the platform enterprise network public opinion authenticity detection method based on a time sequence knowledge graph according to any one of claims 1 to 3.

Images