Transfer learning-based intention topic detection method and device and computer readable storage medium

By balancing the adaptation of distributed adaptive features and building user network structure features, combined with the transfer learning training model, the data imbalance problem in the existing technology is solved, and the accuracy and performance of native topics and derivative topic intention topic detection is improved.

CN118964592AActive Publication Date: 2024-11-15CHONGQING UNIV OF POSTS & TELECOMM
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202410962710.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-18
Publication Date
2024-11-15
Estimated Expiration
2044-07-18

AI Technical Summary

Technical Problem

The existing intent-based topic detection methods are unable to adapt to native and derivative topics at the same time due to data imbalance, which affects performance and accuracy.

Method used

By obtaining labeled native topic data sets and labelless derived topic data sets, balancing distribution adaptive feature adaptation is carried out, and the user network structure characteristics in the topic propagation network are constructed using the SDNE algorithm, combined with transfer learning training intention-based topic detection Model.

Benefits of technology

It effectively reduces the edge distribution differences between native topics and derivative topic data, and improves the accuracy and performance of the model's identification of intent topics.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118964592A_ABST
    Figure CN118964592A_ABST
Patent Text Reader

Abstract

The invention relates to an intention type topic detection method and device based on transfer learning and a computer readable storage medium. The intention type topic detection method comprises the steps that a labeled native topic data set and an unlabeled derivative topic data set are acquired and preprocessed; feature adaptation is carried out on the derivative topic data and the original topic data through balance distribution self-adaption; utilizing an SDNE algorithm to construct network structure features of users under the corresponding topics according to the topic propagation networks of the original topics and the derivative topics; training an intention type topic detection model by utilizing transfer learning according to the original topic data and the derivative topic data after feature adaptation and the structural features of the users under the original topic and the derivative topic; and identifying the derivative topic data through the trained intention type topic detection model. The performance and accuracy of the model are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of social network technology applications, and in particular, relates to an intentional topic detection method, device and computer-readable storage medium based on transfer learning. Background Art

[0002] In the current social media era, users widely discuss and share various topics on online platforms, including some intentional topics. Intentional topics refer to topics with specific purposes or intentions that people discuss and share on social media or other information dissemination platforms. These topics are intended to mislead users, promote specific views, spread false information, or achieve other specific purposes. Compared with ordinary discussion topics, intentional topics have the characteristics of hiding and deceiving, and their purpose is to influence the audience's thinking, behavior or beliefs. These intentional topics will cause inconvenience and distress to users, and may also have a negative impact on social order and public safety.

[0003] As information spreads in social networks, a hot topic may spawn one or more derivative topics. Derivative topics are developed from the original topic, usually by expanding, segmenting or extending the original topic. Derivative topics may explore certain aspects of the original topic in more detail and in more depth, or they may be related to the original topic but not the core. Different users have different behaviors in derivative topics and original topics due to factors such as interests, inconsistent data space distribution, and domain adaptation problems.

[0004] In summary, the existing intent-based topic detection methods have the following technical problems: The problem of sparse effective data in topics. Intent-based topics often contain a large amount of invalid data such as irrelevant data, abnormal data, and noise data. This leads to an unbalanced category distribution, which makes the samples of different categories in the original topics and derived topics unevenly distributed, thus affecting the model's inability to adapt to the intent-based topic detection of both original topics and derived topics at the same time, resulting in low performance and accuracy of the model. Summary of the invention

[0005] In order to solve the problems existing in the background technology, one aspect of the present invention provides an intention-based topic detection method based on transfer learning, comprising:

[0006] S1: Obtain labeled original topic datasets and unlabeled derived topic datasets and perform preprocessing;

[0007] S2: Use balanced distribution adaptation to perform feature adaptation on derived topic data and native topic data;

[0008] S3: Based on the topic propagation network of original topics and derived topics, the SDNE algorithm is used to construct the network structure characteristics of users under the corresponding topics;

[0009] S4: Use transfer learning to train the intention-based topic detection model based on the original topic data and derived topic data after feature adaptation, as well as the network structure characteristics of users under the original topic and derived topic;

[0010] S5: Identify derived topic data through the trained intent-based topic detection model.

[0011] Preferably, the native topic data or derived topic data include but are not limited to: topic content text, comment content text of users under the topic and topic propagation network; the topic propagation network includes: comment relationship network between users under the topic and attention relationship network between users.

[0012] Preferably, step S3 comprises: using the SDNE algorithm to perform representation learning on the comment relationship network and the attention relationship network respectively to extract the comment structure feature W of the user node i The structural characteristics of the relationship between the user and i ; According to the user's theoretical structure characteristics W i The structural characteristics of the relationship between the user and i Get the user's network structure feature u i =μ1×W i +μ2×R i , μ1 and μ2 represent weight parameters.

[0013] Preferably, preprocessing the native topic data and derived topic data includes: segmenting and tagging the topic content text of the native topic data and derived topic data, and the comment content text of the users under the topic, removing useless words and stop words, obtaining candidate keywords for the topic, using the TF-IDF algorithm to calculate the weight of each candidate keyword, extracting keywords with weights greater than a set threshold as the central keywords of the topic, obtaining the central keyword sequence of the native topic and the derived topic, and then using the Doc2vec algorithm to learn the central keyword sequence, and the output feature vectors are: native topic content feature vector, derived topic content feature vector, native topic user content feature vector and derived topic user content feature vector.

[0014] Preferably, the optimization goal of feature adaptation of derived topic data and native topic data using balanced distribution adaptation is expressed as:

[0015]

[0016] sA T XHX T A=I,0≤u≤1

[0017] Among them, tr() represents the trace of the matrix, A represents the feature transformation matrix, X represents the data of the original topic and the derived topic; C represents the number of sample categories; λ represents the regularization parameter; H represents the center matrix; M c represents the maximum mean difference matrix between classes; M0 represents the maximum mean difference matrix between native topics and derived topics; T represents matrix transpose; u represents the balance factor, which is used to adjust the weight of marginal distribution difference and conditional distribution difference in the adaptation process; ||A|| F represents the Frobenius norm of the matrix A; I represents the identity matrix.

[0018] Preferably, the maximum mean difference matrix of the original topic and the derived topic includes:

[0019]

[0020] Among them, (M o ) ij represents the jth element in the i-th row of M0, n s and n t Represents the number of samples of original topic and derived topic respectively; D s and D t Respectively represent the original topic and the derived topic; x i and x j Represent the i-th data and j-th data in X respectively.

[0021] Preferably, the maximum mean difference matrix M c include:

[0022]

[0023] Among them, (M c ) ij Indicates M c The jth element in the i-th row of and Respectively represent the number of samples of the c-th original topic and derived topic; and They represent the c-th original topic and derived topic respectively.

[0024] Preferably, the intention-based topic detection model includes a feature extractor, a label classifier and a domain discriminator; the input features of the feature extractor are the native topic data or derived topic data after feature adaptation, and the network structure features of users under the native topic or derived topic; the output of the feature extractor is the extracted domain-invariant features of the native topic data and the derived topic data; the domain discriminator determines whether the extracted domain-invariant features are from the native topic or the derived topic; the label classifier classifies the extracted domain-invariant features to obtain the intention type of the topic.

[0025] Another aspect of the present invention provides an intention-based topic detection device based on transfer learning, comprising a processor and a memory; the memory is used to store computer programs; the processor is connected to the memory, and is used to execute the computer programs stored in the memory, so that the intention-based topic detection device based on transfer learning executes the intention-based topic detection method based on transfer learning.

[0026] Another aspect of the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the method for detecting intentional topics based on transfer learning.

[0027] The present invention has at least the following beneficial effects

[0028] The present invention performs feature adaptation on derived topic data and native topic data through balanced distribution adaptation to reduce the marginal distribution difference and conditional distribution difference between the native topic data and the derived topic data, so that the target model can accurately identify the topic intent types of the native topic and the derived topic at the same time. At the same time, the present invention extracts the network structure characteristics of the user based on the topic propagation network of the native topic data and the derived topic data, and performs transfer learning on the intent topic detection model in combination with the feature-adapted native topic data and the derived topic data, so that the intent topic recognition model can be simultaneously applicable to the intent detection of native topics and derived topics, thereby improving the performance and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0030] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.

[0031] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0032] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.

[0033] See also Figure 1 One aspect of the present invention provides an intentional topic detection method based on transfer learning, comprising:

[0034] S1: Obtain labeled original topic datasets and unlabeled derived topic datasets and perform preprocessing;

[0035] S2: Use balanced distribution adaptation to perform feature adaptation on derived topic data and native topic data;

[0036] S3: Based on the topic propagation network of original topics and derived topics, the SDNE algorithm is used to construct the network structure characteristics of users under the corresponding topics;

[0037] S4: Use transfer learning to train the intention-based topic detection model based on the original topic data and derived topic data after feature adaptation, as well as the network structure characteristics of users under the original topic and derived topic;

[0038] S5: Identify derived topic data through the trained intent-based topic detection model.

[0039] In this embodiment, a labeled native topic data set and an unlabeled derived topic data set are obtained through the API interface provided by the social platform, and the native topic data or derived topic data include but are not limited to: topic content text, comment content text of users under the topic, and topic propagation network; the topic propagation network includes: a comment relationship network between users under the topic and a follow-up relationship network between users, etc., wherein the nodes in the comment relationship network are represented as user nodes, and the relationship between the nodes represents whether there is a comment relationship between the users, and the nodes in the follow-up relationship network are represented as user nodes, and the relationship between the nodes represents the friend relationship and / or follow-up relationship between the users.

[0040] Preferably, the native topic data or derived topic data include but are not limited to: topic content text, comment content text of users under the topic and topic propagation network; the topic propagation network includes: comment relationship network between users under the topic and attention relationship network between users.

[0041] Preferably, step S3 comprises: using the SDNE algorithm to perform representation learning on the comment relationship network and the attention relationship network respectively to extract the comment structure feature W of the user node i The structural characteristics of the relationship between the user and i ; According to the user's theoretical structure characteristics W i The structural characteristics of the relationship between the user and i Get the user's network structure feature u i =μ1×W i +μ2×R i , μ1 and μ2 represent weight parameters.

[0042] In this embodiment, the SDNE algorithm is used to perform representation learning on the comment relationship network and the attention relationship network respectively, including: SDNE uses a semi-supervised deep model of an opportunity autoencoder to perform network embedding, where the input of the model is the adjacency matrix S of the i-th user node i , S={S1,S2,...,S n} is the adjacency matrix of the network, and the output is the reconstructed adjacency matrix x i The error between the input and reconstruction matrices is continuously reduced until it is less than the set threshold, and the global structural features are obtained; the second-order similarity loss function of SDNE is defined as: Among them, ⊙ represents the Hadamard product, If S i,j = 0 then b i,j =1, otherwise b i,j =β>1. For the output features of each encoding layer of the autoencoder is the local feature of node i; k=1,2,...,K, K represents the number of encoding layers of the autoencoder; W (k) represents the weight matrix of the kth encoding layer of the autoencoder; b (k) The bias term of the kth encoding layer of the autoencoder; the first-order similarity loss function of SDNE is in, represents the output representation of node i at the kth encoding layer; S i,j Indicates whether nodes i and j are connected. If there is an edge, it is 1, otherwise it is 0. The minimum optimization objective function after the combination is: L min =L 2nd +λL 1st +vL reg, where L reg represents the regularization term, λ is the parameter that controls the first-order loss; v represents the parameter that controls the regularization term; finally, the model is trained to obtain the minimum L min After that, the output is used as the user's comment structure feature W i The structural characteristics of the relationship between the user and i ; According to the user's theoretical structure characteristics W i The structural characteristics of the relationship between the user and i Get the user's network structure feature u i =μ1×W i +μ2×R i , μ1 and μ2 represent weight parameters; the structural feature of user comments W i Or the user's relational structure feature R i It consists of the output representations of all encoding layers of the autoencoder.

[0043] Preferably, preprocessing the native topic data and derived topic data includes: segmenting and tagging the topic content text of the native topic data and derived topic data, and the comment content text of the users under the topic, removing useless words and stop words, obtaining candidate keywords for the topic, using the TF-IDF algorithm to calculate the weight of each candidate keyword, extracting keywords with weights greater than a set threshold as the central keywords of the topic, obtaining the central keyword sequence of the native topic and the derived topic, and then using the Doc2vec algorithm to learn the central keyword sequence, and the output feature vectors are: native topic content feature vector, derived topic content feature vector, native topic user content feature vector and derived topic user content feature vector.

[0044] Preferably, the optimization goal of feature adaptation of derived topic data and native topic data using balanced distribution adaptation is expressed as:

[0045]

[0046] sA T XHX T A=I,0≤u≤1

[0047] Among them, tr() represents the trace of the matrix, A represents the feature transformation matrix, X represents the data of the original topic and the derived topic; C represents the number of sample categories; λ represents the regularization parameter; H represents the center matrix; M c represents the maximum mean difference matrix between classes; M0 represents the maximum mean difference matrix between native topics and derived topics; T represents matrix transpose; u represents the balance factor, which is used to adjust the weight of marginal distribution difference and conditional distribution difference in the adaptation process; ||A|| F represents the Frobenius norm of the matrix A; I represents the identity matrix.

[0048] In this embodiment, balanced distribution adaptation is expressed as finding a transformation W so that the difference between the marginal distribution and conditional distribution of the original topic and the derived topic data space after the transformation is significantly reduced. The original topic content feature vector T is defined as S , derived topic content feature vector T D , Native topic user content feature vector U S and the derived topic user content feature vector U T ; Assume X S =X T , Y S =Y T , P(X S )≠P(X T ), Q(Y S |X S )≠Q(Y T |X T ), find a transformation A that can significantly reduce the difference in marginal distribution and conditional distribution between the original topic and the derived topic data space after the transformation. According to the feature vectors of topic content and user content obtained above, assuming that the original topic data X S =[T S ,U S ], derivative topic X T =[T T ,U T ], then the source domain D S ={X S ,P(X s )},X s ∈X S , where the source domain task is T S = {Y s ,f(X s )},Y s ∈Y S ,f(X s ) represents the classifier on the source domain, and the target domain P(X S ) represents the target domain D T ={X T ,P(X t )},X t ∈X T , P(X t ) represents the classifier on the target domain, which is composed of X T Predict Y T .

[0049] By maximizing the divergence, introducing the balance factor u, and continuously adjusting the value of u, the weights of the edge distribution difference and the conditional distribution difference in the adaptation process are adjusted. The kernel function method is used to simplify and obtain the final optimization goal:

[0050]

[0051] sA T XHX T A=I,0≤u≤1

[0052] Finally, the two distances are combined to obtain an overall optimization goal:

[0053]

[0054] Where Φ is the Lagrange multiplier.

[0055] Preferably, the maximum mean difference matrix of the original topic and the derived topic includes:

[0056]

[0057] Among them, (M o ) ij represents the jth element in the i-th row of M0, n s and n t Represents the number of samples of original topic and derived topic respectively; D s and D t Represents original topic and derived topic respectively; x i and x j Represent the i-th data and j-th data in X respectively.

[0058] Preferably, the maximum mean difference matrix M c include:

[0059]

[0060] Among them, (M c ) ij Indicates M c The jth element in the i-th row of and Respectively represent the number of samples of the c-th original topic and derived topic; and They represent the c-th original topic and derived topic respectively.

[0061] Preferably, the intention-based topic detection model includes a feature extractor, a label classifier and a domain discriminator; the input features of the feature extractor are the native topic data or derived topic data after feature adaptation, and the network structure features of users under the native topic or derived topic; the output of the feature extractor is the extracted domain-invariant features of the native topic data and the derived topic data; the domain discriminator determines whether the extracted domain-invariant features are from the native topic or the derived topic; the label classifier classifies the extracted domain-invariant features to obtain the intention type of the topic.

[0062] In this embodiment, the intention-based topic detection model includes a feature extractor, a label classifier, and a domain discriminator; let the native topic-related feature input by the feature extractor be f i s, the label of the original topic is Among them, the feature f i s Including: the original topic data after feature adaptation and the network structure characteristics of users under the original topic; similarly, the derived topic related features input by the feature extractor are f i T , where the feature f i s Including: derived topic data after feature adaptation and network structure characteristics of users under derived topics;

[0063] The feature extractor, label classifier, and domain discriminator are all multi-layer composite functions. For the feature extractor, its output is expressed as:

[0064] G f (x i ;w,b)=sigmoid(wx i +b)

[0065] Among them, x i represents the i-th sample of the input, w represents the weight term of the feature extractor, b represents the bias term of the feature extractor, and sigmoid() represents the activation function; G f (x i ; w, b) represents the extracted inter-domain invariant features;

[0066] For the label classifier, its output is expressed as:

[0067] G y (G f (x i ),v,c)=sigmoid(vG f (x i )+c)

[0068] Among them, v represents the weight term of the label classifier, c represents the bias term of the label classifier; G y(G f (x i ),v,c) represents the predicted label; the loss of the label classifier is L y (G y (G f (x i )),y i )(Abbreviated as L y ) is expressed as:

[0069] L y (G y (G f (x i )),y i )=-logG y (G f (x i ))×y i

[0070] Among them, y i Represents the true label of the sample;

[0071] The domain discriminator shares input features with the label classifier, and the output is the source of the sample, that is, whether the sample comes from the original topic or the derived topic. The domain discriminator loss is:

[0072] L d (G d (G f (x i )),d i )=-d i logG d (G f (x i )-(1-d i )logG d (G f (x i )

[0073] Among them, d i is the domain category label of the i-th sample, and the domain discriminator loss L d (G d (G f (x i )),d i ) is abbreviated as L d ; The feature extractor minimizes the error of the label classifier and continuously improves the classification of the features. At the same time, the feature extractor maximizes the classification error of the domain discriminator, making the features domain invariant. In order to minimize the overall loss function, a gradient reversal layer is added between the feature extractor and the domain discriminator, which remains unchanged during forward propagation and is automatically reversed during back propagation. The total loss function L = L y -ηLd , η represents the regularization parameter of the adversarial transfer learning model; The label classification loss and domain classification loss synergistically optimize the feature extractor to find domain-invariant features so that the training model of the original topic can be transferred to the derived topic and the intention of the derived topic can be predicted.

[0074] In summary, the present invention performs feature adaptation on derived topic data and native topic data through balanced distribution adaptation to reduce the marginal distribution difference and conditional distribution difference between native topic data and derived topic data, so that the target model can accurately identify the topic intention types of native topics and derived topics at the same time. At the same time, the present invention extracts the network structure characteristics of users based on the topic propagation network of native topic data and derived topic data, and performs transfer learning on the intent topic detection model in combination with the feature-adapted native topic data and derived topic data, so that the intent topic recognition model can be simultaneously applicable to the intent detection of native topics and derived topics, thereby improving the performance and accuracy of the model.

[0075] Another aspect of the present invention provides an intention-based topic detection device based on transfer learning, comprising a processor and a memory; the memory is used to store computer programs; the processor is connected to the memory, and is used to execute the computer programs stored in the memory, so that the intention-based topic detection device based on transfer learning executes the intention-based topic detection method based on transfer learning.

[0076] Another aspect of the present invention provides a computer-readable storage medium storing a program, which, when executed by a processor, implements the method for detecting intentional topics based on transfer learning.

[0077] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0078] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0079] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.

Claims

1. A method for detecting intentional topics based on transfer learning, characterized in that: include: S1: Obtain labeled original topic datasets and unlabeled derived topic datasets and perform preprocessing; S2: Use balanced distribution adaptation to perform feature adaptation on derived topic data and native topic data; S3: Based on the topic propagation network of original topics and derived topics, the SDNE algorithm is used to construct the network structure characteristics of users under the corresponding topic; S4: Use transfer learning to train the intention-based topic detection model based on the original topic data and derived topic data after feature adaptation, as well as the network structure characteristics of users under the original topic and derived topic; S5: Identify derived topic data through the trained intent-based topic detection model.

2. According to the method for detecting intentional topics based on transfer learning in claim 1, it is characterized in that: The original topic data or derived topic data include but are not limited to: topic content text, user comment content text under the topic, and topic propagation network; The topic propagation network includes: a comment relationship network between users under a topic and an attention relationship network between users.

3. According to claim 2, the method for detecting intentional topics based on transfer learning is characterized in that: The step S3 includes: using the SDNE algorithm to perform representation learning on the comment relationship network and the attention relationship network respectively to extract the comment structure feature W of the user node i The structural characteristics of the relationship between the user and i ; According to the user's theoretical structure characteristics W i The structural characteristics of the relationship between the user and i Get the user's network structure feature u i =μ1×W i +μ2×R i , μ1 and μ2 represent weight parameters.

4. According to claim 2, the method for detecting intentional topics based on transfer learning is characterized in that: Preprocessing of native topic data and derived topic data includes: segmenting and part-of-speech tagging the topic content text of the native topic data and derived topic data, and the comment content text of users under the topic, removing useless words and stop words, obtaining candidate keywords for the topic, using the TF-IDF algorithm to calculate the weight of each candidate keyword, extracting keywords with weights greater than a set threshold as the central keywords of the topic, obtaining the central keyword sequence of the native topic and derived topic, and then using the Doc2vec algorithm to learn the central keyword sequence, and the output feature vectors are: native topic content feature vector, derived topic content feature vector, native topic user content feature vector and derived topic user content feature vector.

5. According to the method for detecting intentional topics based on transfer learning according to claim 4, it is characterized in that: The optimization goal of feature adaptation of derived topic data and native topic data using balanced distribution adaptation is expressed as: s.t.A T XHX T A=I,0≤u≤1 Among them, tr() represents the trace of the matrix, A represents the feature transformation matrix, X represents the data of the original topic and the derived topic; C represents the number of sample categories; λ represents the regularization parameter; H represents the center matrix; M c represents the maximum mean difference matrix between classes; M0 represents the maximum mean difference matrix between native topics and derived topics; T represents matrix transpose; u represents the balance factor, which is used to adjust the weight of marginal distribution difference and conditional distribution difference in the adaptation process; ||A|| F represents the Frobenius norm of the matrix A; I represents the identity matrix.

6. The method for detecting intentional topics based on transfer learning according to claim 5, characterized in that: The maximum mean difference matrix of the original topic and the derived topic includes: Among them, (M o ) ij represents the jth element in the i-th row of M0, n s and n t Represents the number of samples of original topic and derived topic respectively; D s and D t Represents original topic and derived topic respectively; x i and x j Represent the i-th data and j-th data in X respectively.

7. The method for detecting intentional topics based on transfer learning according to claim 5, characterized in that: The maximum mean difference matrix M c include: Among them, (M c ) ij Indicates M c The jth element in the i-th row of and Respectively represent the number of samples of the c-th original topic and derived topic; and They represent the c-th original topic and derived topic respectively.

8. The method for detecting intentional topics based on transfer learning according to claim 1, characterized in that: The intention-based topic detection model includes a feature extractor, a label classifier and a domain discriminator; the input features of the feature extractor are the original topic data or the derived topic data after feature adaptation, and the structural features of the users under the original topic or the derived topic; the feature The output of the extractor is the domain-invariant features of the extracted original topic data and derived topic data; The domain discriminator uses the extracted domain-invariant features to determine whether it comes from an original topic or a derived topic; the label classifier classifies the extracted domain-invariant features to obtain the intent type of the topic.

9. An intentional topic detection device based on transfer learning, characterized in that: It includes a processor and a memory; the memory is used to store computer programs; the processor is connected to the memory and is used to execute the computer programs stored in the memory, so that the intention-based topic detection device based on transfer learning executes the intention-based topic detection method based on transfer learning described in any one of claims 1-8.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, an intention-based topic detection method based on transfer learning as described in any one of claims 1 to 8 is implemented.

Citation Information

Patent Citations

  • Social network derived topic propagation prediction method based on transfer learning

    CN115470984A

  • Topic propagation prediction method based on topic association

    CN115712772A

  • Key element influence discovery method based on native-derived topic transfer learning

    CN115757518A