A topic model updating method and system, a storage medium and a server

Abstract

The embodiment of the application discloses a kind of theme model updating method, system and storage medium and server, apply to the information processing technical field based on artificial intelligence.Theme model system will obtain the first label semantic feature and the second label semantic feature corresponding respectively to multiple old theme labels in the first theme model and multiple new theme models in the second theme model, and based on the first label semantic feature and the second label semantic label, mapping relationship is established between old theme label and new theme label, and then the old theme label in the first theme model is updated based on the mapping relationship.The updating of the first theme model existing in the system is realized automatically, the efficiency of the theme model is improved, and the theme model with larger dimensionality can also be updated, and the updating of the first theme model is not limited by the theme model acquisition method.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence-based information processing technology, in particular to a topic model updating method and system, a storage medium and a server. BACKGROUND

[0002] In the personalized recommendation scenario, the information stream article topic label is an important characterization dimension of the content understanding of the information stream article. Generally, on the article content understanding side, each article corresponds to a corresponding topic label, and on the user side, each user is also marked with interest in the topic label, so that the user's interested articles are finally recommended through topic label recall, rough sorting and fine sorting.

[0003] Generally, a certain method is used to construct a topic model. If the topic model is not updated after being constructed, the consequence is that as time evolves, the information stream article changes, and many new articles and new hot words appear. This part of information usually has poor recognition effect in the fixed topic model. Further, the recognition effect of the online topic model is continuously reduced over time, affecting the final recommendation accuracy and user experience.

[0004] In the prior art, the topic model can be updated by manual review. Generally, it is only suitable for small topic models in the topic dimension, such as 200-dimensional topics and 500-dimensional topics. For fine-grained topics such as 10,000-dimensional, 50,000-dimensional and 100,000-dimensional topics, the manual evaluation method is time-consuming and laborious. If all topics are evaluated manually, the cost and price are extremely high. In the information stream scenario, it is usually a fine-grained multi-dimensional topic model, so it is difficult to manually update and evaluate all dimensions of the topic model each time the topic model is updated. SUMMARY

[0005] The embodiments of the present application provide a topic model updating method, system, storage medium and server, which automatically updates the topic model.

[0006] The embodiments of the present application provide a topic model updating method, system, storage medium and server, which automatically updates the topic model.

[0007] Obtain first label semantic features corresponding to a plurality of old topic labels in a first topic model respectively; the first label semantic features are used to describe semantic information of a topic represented by the corresponding old topic label;

[0008] Obtain second label semantic features corresponding to a plurality of new topic labels in a second topic model respectively; the second label semantic features are used to describe semantic information of a topic represented by the corresponding new topic label;

[0009] According to the first label semantic feature and the second label semantic feature, a mapping relationship is established between the plurality of old topic labels and the plurality of new topic labels.

[0010] According to the mapping relationship, the old topic labels in the first topic model are updated.

[0011] Another aspect of the embodiment of the present application provides a topic model system, comprising:

[0012] A first feature acquisition unit is configured to acquire first label semantic features corresponding to a plurality of old topic labels in a first topic model respectively; the first label semantic features are used to describe semantic information of topics represented by the corresponding old topic labels;

[0013] A second feature acquisition unit is configured to acquire second label semantic features corresponding to a plurality of new topic labels in a second topic model respectively; the second label semantic features are used to describe semantic information of topics represented by the corresponding new topic labels;

[0014] A mapping unit is configured to establish a mapping relationship between the plurality of old topic labels and the plurality of new topic labels according to the first label semantic features and the second label semantic features.

[0015] An updating unit is configured to update the old topic labels in the first topic model according to the mapping relationship.

[0016] Another aspect of the embodiment of the present application also provides a computer readable storage medium, which stores a plurality of computer programs, the computer programs being adapted to be loaded and executed by a processor to implement the updating method of the topic model according to the aspect of the embodiment of the present application.

[0017] Another aspect of the embodiment of the present application also provides a server, comprising a processor and a memory.

[0018] The memory is configured to store a plurality of computer programs, the computer programs being used to be loaded and executed by the processor to implement the updating method of the topic model according to the aspect of the embodiment of the present application; and the processor is configured to implement each of the plurality of computer programs.

[0019] It can be seen that in the method of the embodiment, the topic model system obtains the first label semantic features and the second label semantic features corresponding to the plurality of old topic labels in the first topic model and the plurality of new topic labels in the second topic model respectively, and establishes a mapping relationship between the old topic labels and the new topic labels based on the first label semantic features and the second label semantic features, and then updates the old topic labels in the first topic model based on the mapping relationship. The method realizes automatic updating of the existing first topic model in the system, improves the efficiency of the topic model, and can also be applied to updating a topic model with a large dimension; and since the label semantic features describing semantics are mainly used in the process of establishing the mapping relationship, the semantics of a topic does not change with the change of the topic model acquisition method, so the mapping relationship established is relatively accurate, and the updating of the first topic model is not limited by the topic model acquisition method. BRIEF DESCRIPTION OF DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained from these drawings without creative labor.

[0021] Figure 1 is a schematic diagram of a topic model updating method provided by an embodiment of the present application;

[0022] Figure 2 is a flowchart of a topic model updating method provided by an embodiment of the present application;

[0023] Figure 3 is a schematic diagram of establishing a mapping relationship in an embodiment of the present application;

[0024] Figure 4 is a flowchart of a topic model updating method provided by an application embodiment of the present application;

[0025] Figure 5 is a schematic diagram of a distributed system to which a topic model updating method in another application embodiment of the present application is applied;

[0026] Figure 6 is a schematic diagram of a block structure in another application embodiment of the present application;

[0027] Figure 7 is a logical structure schematic diagram of a topic model system provided by an embodiment of the present application;

[0028] Figure 8 is a logical structure schematic diagram of a server provided by an embodiment of the present application. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be apparently and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without any creative effort belong to the protection scope of the present application.

[0030] The terms "first", "second", "third", "fourth" and the like (if any) in the description, claims, and drawings of the present application are used to distinguish similar objects, and do not necessarily indicate a particular order or a sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product, or device comprising a series of steps or units does not necessarily have to be limited to the clearly listed steps or units, but can include other steps or units that are not clearly listed or inherent to the process, method, product, or device.

[0031] The subject model updating method provided by the embodiments of the present application can update the old version of the subject model, so as to avoid the problem of inaccuracy of the old version of the subject model caused by the emergence of new information flow, and avoid the problem of poor effect when the information recommendation and other applications are performed based on the old version of the subject model. Specifically, as shown in the figure, the subject model system in the embodiments of the present application updates the old version of the subject model according to the following steps: Figure 1

[0032] The first label semantic features corresponding to the plurality of old topic labels in the first subject model are obtained, and the first label semantic features are used to describe the semantic information of the topics represented by the corresponding old topic labels. The second label semantic features corresponding to the plurality of new topic labels in the second subject model are obtained, and the second label semantic features are used to describe the semantic information of the topics represented by the corresponding new topic labels. The mapping relationship between the plurality of old topic labels and the plurality of new topic labels is established according to the first label semantic features and the second label semantic features. The old topic labels in the first subject model are updated according to the mapping relationship.

[0033] ​In a specific implementation, the subject model system can be applied to a server of any application system (such as a video, audio, multimedia, travel, refueling, charging, etc. application) or a third-party system independent of the server, and can also be applied to, but is not limited to, a user terminal such as a mobile phone, a computer, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, etc. The specific application of the subject model system is not limited herein.

[0034] In this way, the updating of the existing first subject model in the system is automatically implemented, the efficiency of the subject model is improved, and the updating of the subject model with a large dimension can also be applied. Since the label semantic features describing semantics are mainly used in the process of establishing the mapping relationship, the semantics of a subject does not change with the change of the subject model acquisition mode, and therefore the established mapping relationship is relatively accurate, so that the updating of the first subject model is not limited by the subject model acquisition mode.

[0035] The embodiment of the application provides a subject model updating method, which is mainly a method executed by a subject model system, and a flowchart is shown in Figure 2 The method comprises the following steps.

[0036] In step 101, first label semantic features corresponding to a plurality of old subject labels in a first subject model are acquired. The first label semantic features are used to describe semantic information of a subject represented by the corresponding old subject label.

[0037] It can be understood that in the subject model system, a subject model corresponding to a certain information stream can be acquired, and the acquired subject model can be stored in the subject model system in advance. When the storage time of the stored subject model in the subject model system is greater than a preset time, or a new information stream is stored in the subject model system, the subject model system can take the stored subject model as a first subject model and initiate the subject model updating method in the embodiment. Alternatively, the subject model system can take the stored subject model in the system as a first subject model according to a certain period and initiate the method in the embodiment.

[0038] The information stream can include information in any format, such as text, video, audio or picture, and can be applied to any application scenario. The subject model can include a plurality of subject labels and a corresponding relationship between each subject label and an information stream, for example, a certain subject label can correspond to a plurality of articles, and each subject label can uniquely identify a subject to which the corresponding information stream belongs. The subject is mainly a subject based on semantics, and different subject labels can also correspond to the same information stream.

[0039] When the topic model system initiates the process of the embodiment, when obtaining the first label semantic feature, specifically, at least one first information stream corresponding to each old topic label in the first topic model can be determined first, the semantic information of the at least one first information stream is obtained, and then the first label semantic feature is obtained according to the semantic information of the at least one first information stream.

[0040] Among them, the at least one first information stream can be the information stream with the highest prediction probability under the old topic label, where the prediction probability refers to the probability of predicting that the topic label corresponding to the first information stream is the corresponding old topic label, and the prediction probability is obtained in the process of storing the first information stream in the old topic model in advance. In this process, an artificial intelligence-based machine learning model (such as a prediction model, etc.) can be used when predicting the first information stream, which can extract features of the first information stream and directly output the probability of the first information stream corresponding to a certain topic label based on the extracted features. The output probability is the prediction probability. If the probability is greater than a certain value, the first information stream belongs to the topic described by the topic label. If the probability is less than a certain value, the first information stream does not belong to the topic described by the topic label. Or when determining the first information stream, one or more information streams under the old topic label can be randomly selected as the first information stream. Or the first information stream can be selected as the one or more information streams under the old topic label with the most user operations (such as the number of user comments, likes, or forwards, etc.). If there are multiple first information streams, the semantic information of each first information stream in the multiple first information streams can be directly fused to obtain the first label semantic feature, such as the weighted average of the semantic vectors of each first information stream. In this calculation process, the weight values corresponding to each first information stream can be determined according to the prediction probability of the first information stream. Specifically, the prediction probabilities of all first information streams can be divided into multiple intervals, and each first information stream corresponding to the prediction probability of each interval is set with a corresponding weight value. Among them, the first information stream with a higher value in the interval of the prediction probability has a larger weight value.

[0041] Wherein, in the process of obtaining the semantic information of the first information stream, if the first information stream is a text format information stream, the text can be segmented first, and then the features of each segmented word in the text can be fused to obtain the semantic information of the first information stream; or, the frequency of each segmented word in the text can be counted, and the features of the segmented words with the highest frequency can be taken as the semantic information of the first information stream. If the first information stream is a non-text format (such as audio or video or picture, etc.) information stream, the first information stream can be first converted into a text format information stream to obtain a converted information stream, the converted information stream can be segmented, and the features of each segmented word in the converted information stream can be fused to obtain the semantic information of the first information stream. Wherein, in the process of converting the text format information stream, for the picture format information stream, the converted information stream can include the text information contained in the picture; for the audio format information stream, the converted information stream includes the text corresponding to the audio; for the video format information stream, the converted information stream can include the text corresponding to the audio contained in the video.

[0042] Further, in order to make the obtained first label semantic features more accurately describe the old topic label, the topic model system can also combine other dimensions of information of the first information stream when obtaining the first label semantic features, such as obtaining the first non-semantic information based on the first information stream, and then obtaining the first label semantic features according to the semantic information and the first non-semantic information of the first information stream.

[0043] Specifically, the topic model system can obtain the first label semantic features by combining the semantic information and the first non-semantic information of the first information stream through a pre-trained feature extraction model; or, the weighted values of the semantic information and the first non-semantic information of the first information stream can be taken as the first label semantic features, etc.

[0044] Wherein, the first non-semantic information is related to the first information stream, and mainly describes the non-semantic aspects related to the first information stream, which can specifically include but is not limited to at least one of the following information: classification information, tag information, quality score of the first information stream, and user information of the user who publishes the first information stream, etc.

[0045] Step 102, obtaining the second label semantic features corresponding to the plurality of new topic labels in the second topic model respectively. Wherein, the second label semantic features are used to describe the semantic information of the topics represented by the corresponding new topic labels.

[0046] The second topic model is mainly obtained from the emerging information flow or obtained by combining the emerging information flow with the stored information flow in the system. The second topic model can be obtained in a manner different from the first topic model. For example, the first topic model is obtained by using a document topic generation model (Latent Dirichlet Allocation, LDA), and the second topic model is obtained by using a pre-trained language representation model (Bidirectional Encoder Representation from Transformers, BERT).

[0047] In obtaining the second label semantic feature, at least one second information flow corresponding to each new topic label in the second topic model can be determined first, semantic information of the at least one second information flow is obtained, and then the second label semantic feature is obtained according to the semantic information of the at least one second information flow.

[0048] The at least one second information flow can be an information flow with the highest prediction probability under the new topic label. If there are multiple second information flows, the semantic information of each second information flow can be fused to obtain the second label semantic feature. For example, a weighted average of the semantic vectors of each second information flow is used as the second label semantic feature.

[0049] Further, the topic model system can also obtain second non-semantic information of the second information flow, and then obtain the second label semantic feature according to the semantic information and the second non-semantic information of the second information flow. The specific method of obtaining the second label semantic feature is similar to the method of obtaining the first label semantic feature, which will not be described here.

[0050] In step 103, a mapping relationship is established between the multiple old topic labels and the multiple new topic labels according to the first label semantic feature and the second label semantic feature.

[0051] Specifically, the first label semantic feature of the old topic label is matched with the second label semantic feature of each new topic label. If the first label semantic feature of a certain old topic label matches the second label semantic feature of a certain new topic label, a mapping relationship is established between the old topic label and the new topic label.

[0052] The first label semantic feature and the second label semantic feature can be matched in various ways. For example, the distance (such as Euclidean distance or cosine distance) between the first label semantic feature and the second label semantic feature can be calculated. If the calculated distance is within a preset range, the old topic label matches the new topic label, otherwise, the old topic label does not match the new topic label.

[0053] For another example, the topic model system can determine whether any new topic label matches another old topic label according to the first label semantic feature, the second label semantic feature, and a preset matching classification model. The preset matching classification model is an artificial intelligence-based machine learning model used to obtain a result of whether the topic labels corresponding to the two label semantic features match each other according to the two label semantic features. The matching classification model can be trained according to a certain training method, and its running logic is stored in the system in advance.

[0054] In the training of the matching classification model, the training samples can include multiple groups of sample features. Each group of sample features includes semantic features corresponding to two sample labels, and annotation information of whether the two sample labels match each other. Thus, the matching classification model can be supervisedly trained by using the training samples. The multiple groups of sample features can include a certain proportion of positive samples and negative samples. In the positive samples, the two sample labels match each other, while in the negative samples, the two sample labels do not match each other.

[0055] In the training of the matching classification model, the training samples can include multiple groups of sample features. Each group of sample features includes semantic features corresponding to two sample labels, and annotation information of whether the two sample labels match each other. Thus, the matching classification model can be supervisedly trained by using the training samples. The multiple groups of sample features can include a certain proportion of positive samples and negative samples. In the positive samples, the two sample labels match each other, while in the negative samples, the two sample labels do not match each other.

[0056] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software level technology. Artificial intelligence basic technology generally includes, such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0057] Machine Learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. Its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0058] For example Figure 3 As shown, the first topic model A includes old topic labels A1, A2, A3 and AN, and the second topic model B includes new topic labels B1, B2, B3 and BN. By matching the first label semantic features of each old topic label with the second label semantic features of each new topic label, the mapping relationship between old topic label A1 and new topic label B3, old topic label A2 and new topic label B1, old topic label A3 and new topic label BN, and old topic label AN and new topic label B2 is established respectively.

[0059] In step 104, according to the mapping relationship, the old topic labels in the first topic model are updated.

[0060] Specifically, if some new topic labels have corresponding old topic labels in the mapping relationship, the new topic labels can be used to replace the old topic labels with mapping relationship. If some other new topic labels do not have corresponding old topic labels in the mapping relationship, the new topic labels can be directly added to the first topic model. Some old topic labels do not have corresponding new topic labels in the mapping relationship, and these old topic labels will not be updated.

[0061] It should be noted that after updating the first topic model, the updated first topic model can be applied to multiple scenarios, such as using the updated first topic model to recommend information streams under each topic label that users are interested in for the user end, or providing information streams under corresponding topic labels to users according to the user's search information.

[0062] It can be seen that in the method of the embodiment, the topic model system obtains the first label semantic features and the second label semantic features respectively corresponding to the plurality of old topic labels in the first topic model and the plurality of new topic labels in the second topic model, and establishes a mapping relationship between the old topic labels and the new topic labels based on the first label semantic features and the second label semantic features, and then updates the old topic labels in the first topic model based on the mapping relationship. The method realizes automatic updating of the existing first topic model in the system, improves the efficiency of the topic model, and can also be applied to updating a topic model with a large dimension; and since the label semantic features describing semantics are mainly used in the process of establishing the mapping relationship, the semantics of a topic does not change with the change of the topic model acquisition method, so the mapping relationship established is relatively accurate, and the updating of the first topic model is not limited by the topic model acquisition method.

[0063] The following describes the method for updating the topic model of the application with a specific application example. As shown in the following table 1, the method of the embodiment can include the following steps: Figure 4

[0064] Step 201: The topic model system initiates updating of the stored first topic model according to a certain period, and first determines at least one first information flow under each topic label in the first topic model.

[0065] Specifically, N first information flows can be determined for each old topic label, such as N articles representing the corresponding old topic label, and the first information flow can be randomly searched from the old topic model stored in the system to find the information flow represented by the corresponding old topic label, or the information flow with the highest prediction probability under the corresponding old topic label can be selected from the old topic model stored in the system, and the first information flow is usually highly related to the semantic information of the old topic label. For example, the determined articles representing the old topic label can be as shown in the following table 1:

[0066]

[0067]

[0068] Table 1

[0069] Step 202: The topic model system determines the semantic information of the first information flow under each old topic label.

[0070] Specifically, the topic model system can use a pre-trained model such as a BERT model to extract the semantic representation vector of each first information flow.

[0071] Step 203: The topic model system obtains the first label semantic features of each old topic label according to the semantic information of the first information flow.

[0072] ​Specifically, the weighted average of semantic information (such as semantic representation vectors) of all first information streams under each old topic label can be taken as the first label semantic feature of the corresponding old topic label. Further, when obtaining the first label semantic feature, the semantic information and non-semantic information (such as quality scores, user information, etc.) of the first information stream can be fused.

[0073] In step 204, the topic model system obtains second label semantic features of each new topic label under a second topic model.

[0074] Specifically, the second topic model can be obtained according to new information streams and a topic model obtained from stored old information streams in the system. The method for obtaining the second label semantic feature can be the same as that shown in steps 201 to 203 described above, which will not be described here. Moreover, steps 201 to 203 described above do not have an absolute sequential relationship, and can be performed simultaneously or sequentially. The example shown in the figure is only one specific application example.

[0075] In step 205, the topic model system matches the first label semantic feature corresponding to each old topic label with the second label semantic feature corresponding to each new topic label, and establishes a mapping relationship between the old topic label corresponding to the first label semantic feature and the new topic label corresponding to the second label semantic feature.

[0076] Specifically, the mapping relationship can be established by various methods, such as calculating the Euclidean distance or cosine distance between the first label semantic feature and the second label semantic feature, and the distance between the new topic label and the old topic label with the latest calculated distance can establish a mapping relationship.

[0077] Alternatively, a pre-trained matching classification model can be called, and the first label semantic feature and the second label semantic feature are input into the matching classification model, and then the probability value of the matching between the corresponding old topic label and the new topic label is output. If the probability value is greater than a preset value, the old topic label and the new topic label are matched.

[0078] In step 206, the topic model system updates the old topic labels in the first topic model according to the mapping relationship established in step 205.

[0079] Specifically, if some new topic labels have corresponding old topic labels in the mapping relationship, the new topic labels can replace the old topic labels with the mapping relationship. If some other new topic labels do not have corresponding old topic labels in the mapping relationship, the new topic labels can be directly added to the first topic model.

[0080] It should be noted that the above updating of the first topic model is an iterative process, and after the first topic model is updated offline, the updated first topic model can be an important part of content understanding and processing of the information stream, and is also an important feature of recall, rough sorting, fine sorting and the like. The request initiated by the user side can be unperceived. For example,

[0081] For example, when the user initiates a topic analysis request for a certain information stream through an application terminal, the topic model system can return the topic distribution information of the information stream in the topic dimension to the application terminal according to the topic analysis request and the updated topic model, and the multiple topic labels with the highest distribution probability can be taken as the topic distribution information of the information stream.

[0082] It can be seen that in the embodiment, the updating of the topic model can be automatically realized based on the representation and matching of the label semantic features, and the model structures of the two topic models are not limited to be the same, and the cost of manual evaluation of all-dimensional topic labels can be significantly reduced, so that the online topic model effect can continuously meet the business needs.

[0083] The following describes the updating method of the topic model in the application with another specific application example. The topic model system in the embodiment of the application is mainly a distributed system 100, which can include a client 300 and multiple nodes 200 (any form of computing device in an access network, such as a server, a user terminal). The client 300 and the node 200 are connected through network communication.

[0084] Taking the distributed system as a blockchain system, referring to Figure 5 is an optional structure schematic diagram of the distributed system 100 provided by the embodiment of the application, which is formed by multiple nodes 200 (any form of computing device in an access network, such as a server, a user terminal) and a client 300. The nodes form a point-to-point (P2P, Peer To Peer) network, and the P2P protocol is an application layer protocol running on a transmission control protocol (TCP, Transmission Control Protocol) protocol. In the distributed system, any machine such as a server or a terminal can join to become a node, and the node includes a hardware layer, an intermediate layer, an operating system layer and an application layer.

[0085] Referring to Figure 5 The functions of each node in the blockchain system are shown, and the functions involved include:

[0086] 1) Routing, a basic function of the node, used to support communication between nodes.

[0087] In addition to the routing function, the node can also have the following functions:

[0088] 2) Application, for deployment in the blockchain, to implement specific business according to actual business needs, record the data related to the implementation function to form record data, carry the digital signature in the record data to represent the source of the task data, send the record data to other nodes in the blockchain system, and add the record data to the temporary block when the other nodes verify the source and integrity of the record data successfully.

[0089] For example, the application implements a business including code implementing a topic model update function, which mainly includes:

[0090] Obtain the first label semantic feature corresponding to each of the plurality of old topic labels in the first topic model; the first label semantic feature is used to describe the semantic information of the topic represented by the corresponding old topic label; obtain the second label semantic feature corresponding to each of the plurality of new topic labels in the second topic model; the second label semantic feature is used to describe the semantic information of the topic represented by the corresponding new topic label; establish a mapping relationship between the plurality of old topic labels and the plurality of new topic labels according to the first label semantic feature and the second label semantic feature; and update the old topic labels in the first topic model according to the mapping relationship.

[0091] 3) Blockchain, including a series of blocks (Block) connected in time sequence according to the time of generation, once a new block is added to the blockchain, it will not be removed, and the block records the record data submitted by the node in the blockchain system.

[0092] Referring to Figure 6 An optional schematic diagram of the block structure provided by the embodiment of the application is shown, each block includes the hash value of the transaction record stored in the block (the hash value of the block) and the hash value of the previous block, and the blocks are connected by the hash values to form a blockchain. In addition, the block can also include information such as the time stamp when the block is generated. The blockchain is essentially a decentralized database, and is a series of data blocks associated using cryptography. Each data block contains relevant information for verifying the validity (anti-fake) of the information and generating the next block.

[0093] The embodiment of the application also provides a topic model system, a structural schematic diagram of which is shown in Figure 7 As shown, the specific structure can include:

[0094] The first feature acquisition unit 10 is configured to acquire first label semantic features corresponding to a plurality of old topic labels in a first topic model, the first label semantic features being used to describe semantic information of topics represented by the corresponding old topic labels.

[0095] The first feature acquisition unit 10 is configured to determine at least one first information stream corresponding to the old topic label, acquire semantic information of the at least one first information stream, and acquire the first label semantic features according to the semantic information of the at least one first information stream.

[0096] In the process of acquiring the semantic information of the at least one first information stream, if the first information stream is a text format information stream, the first feature acquisition unit 10 is configured to perform word segmentation on the first information stream, fuse features of each word in the first information stream to obtain the semantic information of the first information stream; if the first information stream is a non-text format information stream, the first feature acquisition unit 10 is configured to convert the first information stream into a text format information stream, obtain a converted information stream, perform word segmentation on the converted information stream, and fuse features of each word in the converted information stream to obtain the semantic information of the first information stream.

[0097] Further, the first feature acquisition unit 10 is also configured to acquire first non-semantic information of the first information stream, and in the process of acquiring the first label semantic features according to the semantic information of the at least one first information stream, the first feature acquisition unit 10 is configured to acquire the first label semantic features according to the semantic information of the at least one first information stream and the first non-semantic information.

[0098] The second feature acquisition unit 11 is configured to acquire second label semantic features corresponding to a plurality of new topic labels in a second topic model, the second label semantic features being used to describe semantic information of topics represented by the corresponding new topic labels.

[0099] The second feature acquisition unit 11 is configured to determine at least one second information stream corresponding to the new topic label, acquire semantic information of the at least one second information stream, and acquire the second label semantic features according to the semantic information of the at least one second information stream.

[0100] Further, the second feature acquisition unit 11 is also configured to acquire second non-semantic information of the second information stream, and in the process of acquiring the second label semantic features according to the semantic information of the at least one second information stream, the second feature acquisition unit 11 is configured to acquire the second label semantic features according to the semantic information of the at least one second information stream and the second non-semantic information.

[0101] The mapping unit 12 is configured to establish a mapping relationship between the plurality of old topic labels and the plurality of new topic labels according to the first label semantic features obtained by the first feature obtaining unit 10 and the second label semantic features obtained by the second feature obtaining unit 11.

[0102] The mapping unit 12 is specifically configured to establish a mapping relationship between a certain old topic label and a certain new topic label if the first label semantic features of the certain old topic label match the second label semantic features of the certain new topic label.

[0103] The updating unit 13 is configured to update the old topic labels in the first topic model according to the mapping relationship established by the mapping unit 12.

[0104] The updating unit 13 is specifically configured to replace a corresponding old topic label with the new topic label if the new topic label has the corresponding old topic label in the mapping relationship, and add the new topic label to the first topic model if the new topic label does not have the corresponding old topic label in the mapping relationship.

[0105] In the system of the embodiment, the first feature obtaining unit 10 and the second feature obtaining unit 11 can respectively obtain the first label semantic features and the second label semantic features corresponding to the plurality of old topic labels in the first topic model and the plurality of new topic labels in the second topic model, and the mapping unit 12 can establish a mapping relationship between the old topic labels and the new topic labels based on the first label semantic features and the second label semantic features, and then the updating unit 13 can update the old topic labels in the first topic model based on the mapping relationship. The system can automatically update the existing first topic model, improve the efficiency of the topic model, and be applied to updating a topic model with a large dimension. In addition, the mapping relationship is established based on the label semantic features mainly describing semantics, and the semantics of a topic does not change with the change of the topic model obtaining method, so the established mapping relationship is relatively accurate, and the updating of the first topic model is not limited by the topic model obtaining method.

[0106] The embodiment of the application further provides a server, and a structural schematic diagram of the server is as shown in Figure 8As shown, the terminal device can have a large difference due to different configurations or performances, and can include one or more central processing units (CPUs) 20 (for example, one or more processors) and a memory 21, one or more storage media 22 (for example, one or more mass storage devices) storing one or more application programs 221 or data 222. Among them, the memory 21 and the storage medium 22 can be temporary storage or persistent storage. The program stored in the storage medium 22 can include one or more modules (not shown in the figure), each of which can include a series of instruction operations in the server. Further, the central processing unit 20 can be configured to communicate with the storage medium 22 and execute a series of instruction operations in the storage medium 22 on the server.

[0107] Specifically, the application program 221 stored in the storage medium 22 includes an application program for updating the topic model, and the program can include the first feature acquisition unit 10, the second feature acquisition unit 11, the mapping unit 12 and the updating unit 13 in the topic model system described above, which will not be repeated here. Further, the central processing unit 20 can be configured to communicate with the storage medium 22 and execute a series of operations corresponding to the application program for updating the topic model stored in the storage medium 22 on the server.

[0108] The server can also include one or more power supplies 23, one or more wired or wireless network interfaces 24, one or more input / output interfaces 25, and / or one or more operating systems 223, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM, etc.

[0109] The steps performed by the topic model system in the method embodiments described above can be based on the Figure 8 The structure of the server as shown.

[0110] Further, another aspect of the embodiments of the present application also provides a computer readable storage medium storing a plurality of computer programs, the computer programs being adapted to be loaded and executed by a processor to perform the method for updating the topic model as performed by the topic model system described above.

[0111] Another aspect of the embodiments of the present application also provides a server including a processor and a memory;

[0112] The memory is used to store a plurality of computer programs, the computer programs being used to be loaded and executed by the processor to perform the method for updating the topic model as performed by the topic model system described above; and the processor is used to implement each computer program in the plurality of computer programs.

[0113] In addition, according to an aspect of the present application, a computer program product or computer program is provided, which includes computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the computer device executes the subject model updating method provided in the various optional implementation manners.

[0114] A person of ordinary skill in the art can understand that all or part of the steps in the above-mentioned various methods of the embodiments can be completed by a program instructing relevant hardware, and the program can be stored in a computer readable storage medium, which can include a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0115] The subject model updating method, system, storage medium and server provided by the embodiments of the present application are described in detail above, and the principles and implementation manners of the present application are described by applying specific examples; the above embodiment descriptions are only used to help understand the method of the present application and its core idea; meanwhile, for a person of ordinary skill in the art, according to the idea of the present application, the specific implementation manners and application ranges will be changed, and the above descriptions should not be understood as limiting the present application.

Claims

1. A method of updating a topic model, characterized by, The method comprises the following steps: determining at least one first information stream under each old topic label in a first topic model, determining semantic information and first non-semantic information of the at least one first information stream, the first information stream being an information stream with the highest prediction probability obtained when the old topic label is predicted by a prediction model, the first non-semantic information being information of other dimensions of the first information stream except the semantic information, the first information stream being an information stream in a text format or an information stream in a non-text format; obtaining first label semantic features corresponding to respective old topic labels in the first topic model according to the semantic information and the first non-semantic information of the at least one first information stream, the first label semantic features being used to describe semantic information of a topic represented by the respective old topic label, wherein if there are multiple first information streams, the semantic information of each first information stream in the multiple first information streams is fused to obtain the first label semantic features; determining at least one second information stream under each new topic label in a second topic model, determining semantic information and second non-semantic information of the at least one second information stream, the second information stream being an information stream with the highest prediction probability under the new topic label, the second non-semantic information being information of other dimensions of the second information stream except the semantic information; obtaining second label semantic features corresponding to respective new topic labels in the second topic model according to the semantic information and the second non-semantic information of the at least one second information stream, the second label semantic features being used to describe semantic information of a topic represented by the respective new topic label; the first topic model is obtained in a document topic generation model manner, and the second topic model is obtained in a pre-trained language representation model manner; establishing a mapping relationship between multiple old topic labels and multiple new topic labels according to the first label semantic features and the second label semantic features; updating the old topic labels in the first topic model according to the mapping relationship; The determination of the semantic information of the at least one first information stream specifically comprises: if the first information stream is an information stream in a non-text format, converting the first information stream into an information stream in a text format to obtain a converted information stream, performing word segmentation on the converted information stream, and fusing features of each word in the converted information stream to obtain the semantic information of the first information stream.

2. The method of claim 1, wherein, The determination of the semantic information of the at least one first information stream specifically comprises: if the first information stream is an information stream in a text format, performing word segmentation on the first information stream, and fusing features of each word in the first information stream to obtain the semantic information of the first information stream.

3. The method of claim 1, wherein, The method further comprises: obtaining first non-semantic information of the first information stream and second non-semantic information of the second information stream; The obtaining of the first label semantic features according to the semantic information of the at least one first information stream specifically comprises: obtaining the first label semantic features according to the semantic information and the first non-semantic information of the at least one first information stream; The second label semantic feature is obtained according to semantic information of the at least one second information stream, and specifically, the second label semantic feature is obtained according to the semantic information and second non-semantic information of the at least one second information stream.

4. The method according to any one of claims 1 to 3, characterized in that, The mapping relationship is established between the plurality of old theme labels and the plurality of new theme labels according to the first label semantic feature and the second label semantic feature, and specifically, the mapping relationship is established between a certain old theme label and a certain new theme label if the first label semantic feature of the certain old theme label matches the second label semantic feature of the certain new theme label. The old theme label in the first theme model is updated according to the mapping relationship, and specifically, if the new theme label has a corresponding old theme label in the mapping relationship, the corresponding old theme label is replaced by the new theme label; and if the new theme label does not have a corresponding old theme label in the mapping relationship, the new theme label is added to the first theme model.

5. The method according to any one of claims 1 to 3, wherein The old theme label in the first theme model is updated according to the mapping relationship, and specifically, if the new theme label has a corresponding old theme label in the mapping relationship, the corresponding old theme label is replaced by the new theme label; and if the new theme label does not have a corresponding old theme label in the mapping relationship, the new theme label is added to the first theme model. The first feature acquisition unit is configured to determine at least one first information stream under each old theme label in a first theme model, determine semantic information and first non-semantic information of the at least one first information stream, obtain first label semantic features corresponding to respective old theme labels in the first theme model according to the semantic information and the first non-semantic information of the at least one first information stream, and determine the first label semantic features to describe semantic information of themes represented by the respective old theme labels, wherein the first information stream is a text format information stream or a non-text format information stream, the semantic information of each first information stream in a plurality of first information streams is fused to obtain the first label semantic feature, and the first theme model is obtained by using a document theme generation model; The second feature acquisition unit is configured to determine at least one second information stream under each new theme label in a second theme model, determine semantic information and second non-semantic information of the at least one second information stream, obtain second label semantic features corresponding to respective new theme labels in the second theme model according to the semantic information and the second non-semantic information of the at least one second information stream, and determine the second label semantic features to describe semantic information of themes represented by the respective new theme labels, wherein the second information stream is a text format information stream or a non-text format information stream, the semantic information of each second information stream in a plurality of second information streams is fused to obtain the second label semantic feature, and the second theme model is obtained by using a pre-trained language representation model; 6. A topic model system characterized by, The mapping unit is configured to establish a mapping relationship between a plurality of old theme labels and a plurality of new theme labels according to the first label semantic feature and the second label semantic feature. The updating unit is configured to update old theme labels in the first theme model according to the mapping relationship. ​ ​ ​ The first feature acquisition unit is specifically configured to: if the first information stream is a non-text format information stream, convert the first information stream into a text format information stream to obtain a converted information stream, perform word segmentation on the converted information stream, fuse features of each word in the converted information stream, and obtain semantic information of the first information stream.

7. The system of claim 6, wherein, The updating unit is specifically configured to: if the new topic label has a corresponding old topic label in the mapping relationship, replace the corresponding old topic label with the new topic label; and if the new topic label does not have a corresponding old topic label in the mapping relationship, add the new topic label to the first topic model.

8. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a plurality of computer programs, which are adapted to be loaded and executed by the processor to perform the updating method of the topic model according to any one of claims 1 to 5.

9. A server, characterized by The computer readable storage medium stores a plurality of computer programs, which are adapted to be loaded and executed by the processor to perform the updating method of the topic model according to any one of claims 1 to 5. The computer readable storage medium stores a plurality of computer programs, which are adapted to be loaded and executed by the processor to perform the updating method of the topic model according to any one of claims 1 to 5. The computer readable storage medium stores a plurality of computer programs, which are adapted to be loaded and executed by the processor to perform the updating method of the topic model according to any one of claims 1 to 5.

Images