Joint representation fusion method, system, medium, and product for alignment of dual-persona theory

Abstract

The application discloses a dual-personality theory alignment joint representation fusion method, system, medium and product, relates to the technical field of artificial intelligence, and comprises the following steps: obtaining text data of a user within a preset time window, extracting text semantic features, language style features and behavior statistical features of the user and fusing the features, inputting the user feature representation into a big five personality encoder and an MBTI encoder respectively after the user feature representation is constructed, and generating big five personality representation and MBTI representation; based on a preset theory correlation matrix, the big five personality representation and the MBTI representation are cross-theory aligned; the aligned big five personality representation is used as prior information to generate a fusion weight for modulating the aligned MBTI representation, and the aligned big five personality representation and the modulated MBTI representation are fused according to the fusion weight to generate a joint representation; the method aims to establish a joint modeling mechanism, so as to generate a joint representation with integrity and consistency.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a joint representation fusion method, system, medium, and product aligned with dual personality theory. Background Technology

[0002] In recent years, with the widespread application of personality-driven intelligent agent modeling and user profiling technologies in simulation decision-making, virtual interaction and personalized services, how to construct stable and interpretable personality representations from real behavior and text data has become a key issue. Among them, the Big Five personality traits and MBTI characterize individual psychological structures from the perspectives of trait strength and cognitive preference, respectively. However, existing methods usually rely on a single personality theory for modeling, which makes it difficult to simultaneously cover long-term stable traits and contextualized decision-making styles.

[0003] Current personality modeling methods based on single-personality theories or weakly coupled multi-personality composites exhibit systemic flaws in complex cognitive characterization and intelligent agent profiling. The root cause lies in the lack of causal consistency and semantic alignment mechanisms between personality representation structures and psychological theories. Specifically, this manifests in the following aspects: 1. Limited use of personality theories: Existing methods typically rely solely on the Big Five personality traits or the MBTI as the basis for high-level personality modeling. Since the Big Five emphasizes the strength of stable traits, and the MBTI focuses on cognitive and decision-making preferences, a single theory cannot simultaneously cover both types of psychological factors. When a model uses only one theory, the constraints of personality on cognitive behavior are incomplete, leading to instability in the agent's behavioral style and decision-making tendencies.

[0004] 2. Lack of structured alignment mechanism: Psychological research shows that there are stable statistical associations between the Big Five dimensions and MBTI preferences, but current technologies rarely explicitly incorporate these cross-theoretical relationships into model constraints. Models rely solely on data-driven fitting during learning and integration, which can easily deviate from the true psychological structure, resulting in inconsistencies between the two personality representations when expressed uniformly, thus weakening their ability to support complex cognitive modeling.

[0005] 3. Semantic Conflict in Personality: To compensate for the shortcomings of a single personality model, some methods directly concatenate or weightedly fuse the representations of the Big Five and MBTI. However, the dimensional meanings and statistical distributions of the two personality theories are not consistent. Simple concatenation will mix psychological characteristics at different semantic levels in the same representation space, which can easily lead to the repeated expression or mutual cancellation of the same psychological information, thereby weakening the clarity and interpretability of the joint personality vector.

[0006] In summary, existing technologies still have shortcomings in joint personality modeling, mainly in the lack of cross-theoretical alignment and structured constraints in a unified semantic space, which limits the collaborative expressive ability of Big Five and MBTI in complex cognitive modeling. Summary of the Invention

[0007] Based on the technical problems existing in the background technology, this invention proposes a joint representation fusion method, system, medium and product aligned with dual personality theory, aiming to establish a joint modeling mechanism to generate joint representations that are both complete and consistent.

[0008] The proposed dual-personality theory-aligned joint representation fusion method includes: The text data of the user within a preset time window is obtained, and the text semantic features, language style features and behavioral statistical features of the user are extracted and fused. After constructing the user feature representation, they are respectively input into the Big Five Personality Encoder and MBTI Encoder to generate Big Five Personality Representation and MBTI Representation with the same dimension. In the alignment module, based on the theoretical correlation matrix between the Big Five personality traits and the MBTI dimensions pre-set by psychological theory, the Big Five personality traits and the MBTI traits are cross-theoretical aligned to obtain the aligned Big Five personality traits and the aligned MBTI traits. In the representation fusion module, the aligned Big Five personality traits are used as prior information to generate fusion weights for modulating the aligned MBTI representation. The aligned Big Five personality traits and the modulated MBTI representation are then fused according to the fusion weights to generate a joint representation.

[0009] Furthermore, the construction of the user feature representation specifically includes: For each piece of text from the user within a preset time window, a text semantic vector is extracted using a text encoder; Average pooling is performed on all text semantic vectors of the user to obtain a user-level semantic representation; Extract the user's language style feature vector and behavioral statistical feature vector; The user-level semantic representation, language style feature vector, and behavioral statistical feature vector are concatenated to form the user feature representation.

[0010] Furthermore, cross-theoretical alignment is performed on the Big Five personality traits and the MBTI traits, specifically including: On a batch of users, calculate the difference between the Big Five personality traits and the MBTI traits of the same user, and construct user-level consistency constraints; The Big Five personality traits and MBTI traits in the batch of users are input into the corresponding personality classifiers to obtain the Big Five personality prediction results and MBTI prediction results. Calculate the empirical correlation matrix between the Big Five Personality Prediction Results and the MBTI Prediction Results on the batch of users; Based on the difference between the theoretical correlation matrix and the empirical correlation matrix, a cross-theoretical correlation loss is constructed. By combining the user-level consistency constraint and the cross-theoretical correlation loss, an alignment objective function is constructed to align the Big Five personality traits and MBTI representations.

[0011] Furthermore, the alignment objective function The formula is as follows: ; in, For cross-theoretical correlation loss, For user-level consistency constraints, These are the weighting coefficients.

[0012] Furthermore, the generation of joint representations specifically includes: The aligned Big Five personality traits are input into a parameterized gating mapping network and then subjected to nonlinear activation to generate a gating weight vector with the same dimensions as the aligned MBTI traits. The gated weight vector is multiplied element-wise with the aligned MBTI representation to obtain the modulated MBTI representation. The aligned Big Five personality traits are added to the modulated MBTI traits to obtain the joint traits.

[0013] Furthermore, it also includes: The joint representation is input into the downstream task network, and the task loss is constructed based on the supervision signal of the downstream task network. The representation fusion model is trained end-to-end using a joint loss function, which is obtained by weighted combination of the alignment objective function and the task loss. The representation fusion model includes the Big Five Personality Encoder, the MBTI Encoder, the alignment module, the representation fusion module, and the downstream task network.

[0014] Furthermore, the joint loss function The formula is as follows: ; in, To align the balance parameters, Let the personality alignment loss function be... This is the loss function for downstream tasks.

[0015] A computer system includes a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the method described above.

[0016] A computer-readable storage medium storing a plurality of computer programs for being invoked by a processor and executing the method described above.

[0017] A computer program product includes a computer program that is executed by a processor to implement the method described above.

[0018] The advantages of the joint representation fusion method, system, medium, and product aligned with dual personality theories provided by this invention are as follows: It overcomes the inherent differences between the Big Five personality traits and MBTI in terms of dimensional definition and quantification, establishing an intrinsic connection between the two theories at the underlying logical level. It aims to solve the semantic fragmentation and mutual exclusion problems caused by independent modeling or simple superposition in existing technologies, ensuring that the two can produce consistent expressions following psychological principles during joint modeling, rather than being unrelated independent composites. It establishes a scientific fusion mechanism, avoiding information redundancy caused by traditional splicing methods, enabling the trait attributes of the Big Five personality traits and the preference style of MBTI to support and synergize with each other in the same representation. It aims to solve the problems of one-sided characterization by a single theory and the lack of explanatory power due to loose coupling, thereby ensuring that the final generated representation can comprehensively and accurately reflect the individual's joint representation in complex situations. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the structure of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will now be described in detail through specific embodiments. Many specific details are set forth in the following description to provide a thorough understanding of the invention. However, the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0021] like Figure 1 As shown, the joint representation fusion method for aligning dual personality theories proposed in this invention includes: Step 1: Obtain the user's text data within a preset time window, extract the user's text semantic features, language style features, and behavioral statistical features, and fuse them to construct a user feature representation for dual-path personality representation generation; Step 2: Input the user feature representation into the Big Five Personality Encoder and MBTI Encoder respectively to generate Big Five Personality Representations and MBTI Representations with the same dimensions; Step 3: Based on the theoretical correlation matrix between the Big Five personality traits and the various dimensions of MBTI, which is pre-set by psychological theory, perform cross-theoretical alignment on the Big Five personality traits and the MBTI traits to obtain aligned Big Five personality traits and aligned MBTI traits. Step 4: Using the aligned Big Five personality traits as prior information, generate fusion weights for modulating the aligned MBTI representation, and fuse the aligned Big Five personality traits with the modulated MBTI representation according to the fusion weights to generate a joint representation.

[0022] To address the core shortcomings of existing technologies, such as the difficulty in simultaneously considering trait strength and cognitive preferences with single-personality theories, and the redundancy and conflict in expressions caused by the lack of theoretical connection in existing dual-personality fusion methods, the technical objective of this embodiment is to propose a personality representation generation method that uses the Big Five personality traits and MBTI for collaborative modeling. This method aims to abandon the simple feature splicing pattern and establish a joint modeling mechanism that can deeply integrate the advantages of both theoretical sets, thereby generating a dual-personality joint representation (i.e., joint representation) that is both complete and consistent. Specifically, this embodiment focuses on solving the following two core technical problems: 1. How to achieve effective alignment of the Big Five Personality Traits and MBTI at the deep semantic level. This embodiment overcomes the inherent differences between the Big Five Personality Traits and MBTI in terms of dimensional definition and quantification methods, establishing an intrinsic connection between the two theories at the underlying logical level. It aims to solve the semantic fragmentation and mutual exclusion problems caused by independent modeling or simple superposition in existing technologies, ensuring that the two can produce consistent expressions following psychological principles when jointly modeling, rather than being unrelated independent combinations.

[0023] 2. How to construct a deeply complementary joint representation of the Big Five personality traits and MBTI. This embodiment establishes a scientific fusion mechanism that avoids information redundancy caused by traditional splicing methods, enabling the trait attributes of the Big Five personality traits and the preference style of MBTI to support and synergize with each other in the same representation. It aims to solve the problems of one-sided characterization by a single theory and the lack of explanatory power due to loose coupling, thereby ensuring that the final generated representation can comprehensively and accurately reflect the individual's joint representation in complex situations.

[0024] I. Constructing user feature representations; Let the user set be ,in, For the total number of users, each Indicates an individual user. Index for users.

[0025] In the time window Internally, for each user Collect its corpus : ; in, For this user The number of text entries.

[0026] For users Every text posted A text encoder is used. Semantic encoding is performed on it. Specifically, the encoder... Text Mapped to a low-dimensional continuous vector representation: ; in, Indicates user The The text semantic vector of a text.

[0027] It is understood that the core function of the text encoder in this embodiment is to map text into text semantic vectors. The text encoder can be a pre-trained language model (such as BERT, GPT series, etc.), a recurrent neural network and its variants (such as LSTM, GRU, etc.), a convolutional neural network (such as TextCNN), etc.

[0028] Furthermore, in order to acquire users During this time window The overall text semantic representation within the user is obtained by average pooling all semantic representation vectors of the user. : .

[0029] To comprehensively portray a user's personality traits, in addition to user-level semantic representation In addition, this embodiment simultaneously introduces language style feature vectors. With behavioral statistical feature vectors Among them, language style feature vectors This is used to characterize users' language usage habits during text expression, specifically by statistically analyzing the average text length of texts published by users within a preset time window. Vocabulary richness and the use of emotion symbols get.

[0030] .

[0031] Behavioral feature vector This reflects users' behavioral patterns and interaction preferences on the platform, including posting. ,Forward and interaction Statistical characteristics of behaviors: .

[0032] In obtaining the user's user-level semantic representation Language style feature vector and behavioral statistical feature vectors Then, the above features are fused to construct the final user feature representation. : .

[0033] II. Generation of dual personality traits; Based on user feature representation This embodiment further constructs two independent network personality encoders, which are used to model user representations under different personality theory systems. Specifically, user features are represented... Input the Big Five Personality Encoder separately With MBTI encoder We obtained the Big Five personality traits and MBTI traits with the same dimensions: ; in, The Big Five personality traits For MBTI characterization, It is a unified feature dimension.

[0034] At the implementation level, the Big Five Personality Encoder With MBTI encoder Both can be achieved using multilayer fully connected neural networks with similar structures but independent parameters.

[0035] The input layer is used to receive user feature representations. The hidden layer consists of one or more fully connected layers, each followed by a non-linear activation function; the output layer is used to output dimensions... The personality representation vector. In one specific implementation, both encoders use a three-layer fully connected network, with hidden layer dimensions set to 512 and 256 respectively, and output dimensions... Set it to 128.

[0036] III. Representation Alignment Based on the Theory of Dual Personality; In a batch of user sets Above, set user-level consistency constraints for the two representations. : .

[0037] This constraint This is used to reduce the differences in representation of the same user in different personality theory spaces and enhance the consistency of dual-path personality representation.

[0038] Based on this, the dual personality representations (Big Five personality traits and MBTI representation) are input into the corresponding personality classifiers. and The prediction results for users under different personality systems were obtained: ; ; in, For the Big Five personality type classifier, For MBTI classifier, This refers to the prediction results of the Big Five personality traits. Specifically, it represents the user's classification across the dimensions of the Big Five personality traits. This refers to the MBTI prediction result, which is the classification result of the user on the MBTI dimensions.

[0039] To characterize the statistical relationships between different dimensions of personality theory, this embodiment uses a batch of user sets. Calculate the prediction results of the Big Five personality traits. Compared with MBTI prediction results Empirical correlation matrix between ,satisfy: ; in, The Pearson correlation coefficient is used. Indicates the first The five major personality dimensions and the first Empirical correlation matrix between MBTI dimensions.

[0040] This embodiment does not directly apply existing psychological conclusions as fixed rules, but instead uses the correlation between the Big Five personality traits and the various dimensions of MBTI from psychological research as theoretical priors to construct a cross-theoretical correlation matrix. Furthermore, during the training of the representation fusion model, consistency calibration is performed with the empirical correlation matrix obtained from the model prediction, thereby achieving a combination of theoretical priors and data learning results.

[0041] Specifically, let the theoretical correlation matrix be set. for: ; in, This indicates the number of dimensions in the Big Five personality traits. This indicates the number of MBTI dimensions; in this embodiment, it can be set as follows: Take 5. Take 4. Matrix elements This indicates the theoretical correlation strength between the e-th Big Five personality traits and the f-th MBTI dimension.

[0042] Theoretical correlation matrix The construction process includes: (a1) Collect empirical data in psychology; Based on published psychology journal articles, meta-analysis studies, or normative data from standardized personality scales (such as NEO-PI-R and MBTI), obtain the correlation coefficients between each dimension of the Big Five personality traits and each dimension of the MBTI.

[0043] (a2) Determine the values ​​of the matrix elements; The correlation coefficient (usually between -1 and 1) is used directly as... The value can be set, or the correlation can be discretized and graded according to the research conclusions (e.g., strong positive correlation = 0.8, moderate positive correlation = 0.5, weak correlation = 0.1, etc.).

[0044] (a3) Forms a 5×4 matrix; The matrix is ​​constructed as follows: ; Among them, row index These correspond to the five dimensions of the Big Five personality traits (usually in the following order: openness, conscientiousness, extraversion, agreeableness, neuroticism). These correspond to the four dimensions of MBTI.

[0045] matrix It is used to characterize the theoretical relationship between the Big Five personality traits and the various dimensions of MBTI, and to serve as a reference for subsequent cross-theoretical constraints.

[0046] To further constrain the correspondence of dual personality theory at the dimensional level, this embodiment is based on an empirical correlation matrix. correlation matrix with theory Constructing cross-theoretical correlation loss : .

[0047] This cross-theoretical correlation loss This is used to measure the deviation between the experiential relevance learned by the representation fusion model and the prior psychological theories, thereby guiding the dual-path personality representation to maintain consistency with theoretical knowledge at the dimensional level. Thus, the theory relevance matrix... It is not used as the final output, but as a prior constraint in the cross-theoretical alignment process to calibrate the empirical correlation matrix learned by the model. .

[0048] This embodiment does not rely on simple splicing or pure data-driven fitting, but instead uses the Big Five personality traits—MBTI dimensions—as a priori reference and calculates the theoretical correlation matrix through model prediction results. Correlation matrix with experience Consistency calibration is performed to constrain the correspondence between the two personality theories at the dimensional level, thereby achieving stable alignment across the theoretical semantic space.

[0049] Based on this, user-level consistency loss Cross-theoretical correlation loss To perform the union, the overall alignment objective function is defined as: ; Here, is the weighting coefficient, used to balance the influence between user-level consistency constraints and cross-theoretical dimension correlation constraints. Finally, the aligned Big Five personality traits are obtained. and aligned MBTI characterization This provides a consistent and theoretically constrained representation of personality for subsequent tasks.

[0050] IV. Representation Fusion; After obtaining the aligned dual-path personality representation ( and Afterwards, these theories are further integrated to fully leverage the complementary advantages of different personality theories in modeling user characteristics. Specifically, the aligned Big Five personality traits are input into a parameterized gating mapping network and processed by a nonlinear activation function to generate a gating weight vector consistent with the MBTI representation dimensions. The parameterized gating mapping network is a network with a learnable parameter matrix. A fully connected layer with an activation function at its core.

[0051] That is, the aligned Big Five personality traits As stable prior information, it is used to generate the gating weight vector. : ; in, For learnable parameter matrix, For activation functions, such as the Sigmoid activation function. is the gating weight vector, used to characterize the modulating strength of the MBTI representation during the fusion process.

[0052] The gated weight vector is multiplied element-wise with the aligned MBTI representation to obtain the modulated MBTI representation; the aligned Big Five personality traits representation is added to the modulated MBTI representation to obtain the joint representation. This fusion approach preserves the stable characteristics of the Big Five personality traits while incorporating supplementary semantic information provided by the MBTI representation.

[0053] ; in, For the characterization of the modulated MBTI, This is the Hadamard product.

[0054] Joint characterization Input to downstream task network The task loss is defined based on the specific application scenario. For example, in a user interest classification scenario, Used to predict user interest categories Indicates interest category tags; in user behavior prediction scenarios, Used to predict the types of user behavior such as posting, forwarding, commenting, or interacting in the future. This indicates the behavior type label and the corresponding task loss. Cross-entropy loss can be used: ; in, Represents the cross-entropy loss function. This indicates a monitoring signal for downstream tasks.

[0055] Finally, a joint loss function is used to train the representation fusion model end-to-end, and the personality alignment loss and downstream task loss are weighted and combined.

[0056] Finally, a joint loss function is adopted. The personality alignment loss function With downstream task loss function Perform weighted combination: ; in, Alignment balancing parameters are used to balance the impact between personality alignment goals and downstream task optimization goals.

[0057] Understandably, the representation fusion model includes all trainable components from input to output, primarily including: a five-personity encoder. With MBTI encoder Cross-theory alignment module (including user-level consistency loss) and cross-theoretical correlation loss The computational logic); the representation fusion module (containing a learnable parameter matrix for generating fusion weights). and downstream task networks .

[0058] Through the joint loss function End-to-end joint optimization is performed on the representation fusion model to adjust the parameters of the aforementioned trainable components.

[0059] This embodiment combines user-level consistency constraints and the cross-theory correlation loss in a weighted manner to form an end-to-end joint optimization objective. This allows the representation fusion model to learn the cross-theory consistency and dimensional correlation constraints of the dual personality while directly optimizing the overall representation for specific application tasks, thereby improving the effectiveness, stability, and interpretability consistency of the joint representation in real-world application scenarios.

[0060] After training and optimization, the representation fusion model is used to perform dual personality theory alignment. First, user corpus is acquired, and then user feature representation is obtained through step one. This user feature representation is then input into the representation fusion model, and finally, a joint representation is output.

[0061] This embodiment introduces a cross-theoretical alignment mechanism and joint representation fusion method between the Big Five Personality Traits and MBTI, achieving collaborative modeling and consistent expression of dual personality traits in a unified semantic space. This significantly improves the stability and interpretability of personality representations in complex cognitive modeling and agent profiling. Specifically, it has the following main beneficial effects: First, it achieves deep semantic coupling between the two personality theories: the generated joint personality representation contains both the stable trait information of the Big Five personality traits and the cognitive preference information of MBTI, and maintains consistency in the same semantic space, effectively solving the expression conflict problem caused by the fragmented modeling and simple splicing of traditional multi-personality models.

[0062] Second, it ensures the psychological consistency and modeling quality of the joint representation: by introducing cross-theoretical relevance loss and user-level consistency constraint mechanism, the system takes into account the psychological correlation and representation stability between personality dimensions during the fusion process, ensuring that the generated personality vectors are reliable and interpretable at both the theoretical and statistical levels.

[0063] Third, extremely high versatility and scalability: This dual-personality alignment and fusion framework can be flexibly embedded into different application scenarios, and its expressive power can be continuously enhanced as psychological knowledge and data scale expand, providing a stable and unified personality-driven representation.

[0064] Based on the above description of the embodiments, those skilled in the art will understand that the joint representation fusion method, system, medium, and product aligned with dual personality theory described in this embodiment can be implemented in pure software or deployed and run on general-purpose or dedicated computing hardware platforms. Based on this essence, the technical solution of this embodiment can be specifically implemented in the form of a software product containing program instructions. This software product can be stored on various non-volatile storage media or directly deployed as a local or cloud service. The program instructions are used to cause computer devices with processing capabilities—including but not limited to personal computers, server clusters, mobile terminals, or other network devices—to execute the steps described in this embodiment.

[0065] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A joint representation fusion method aligned with dual personality theory, characterized in that, include: The text data of the user within a preset time window is obtained, and the text semantic features, language style features and behavioral statistical features of the user are extracted and fused. After constructing the user feature representation, they are respectively input into the Big Five Personality Encoder and MBTI Encoder to generate Big Five Personality Representation and MBTI Representation with the same dimension. In the alignment module, based on the theoretical correlation matrix between the Big Five personality traits and the MBTI dimensions pre-set by psychological theory, the Big Five personality traits and the MBTI traits are cross-theoretical aligned to obtain the aligned Big Five personality traits and the aligned MBTI traits. In the representation fusion module, the aligned Big Five personality traits are used as prior information to generate fusion weights for modulating the aligned MBTI representation. The aligned Big Five personality traits and the modulated MBTI representation are then fused according to the fusion weights to generate a joint representation. Cross-theoretical alignment of the Big Five personality traits and the MBTI traits is performed, specifically including: On a batch of users, calculate the difference between the Big Five personality traits and the MBTI traits of the same user, and construct user-level consistency constraints; The Big Five personality traits and MBTI traits in the batch of users are input into the corresponding personality classifiers to obtain the Big Five personality prediction results and MBTI prediction results. Calculate the empirical correlation matrix between the Big Five Personality Prediction Results and the MBTI Prediction Results on the batch of users; Based on the difference between the theoretical correlation matrix and the empirical correlation matrix, a cross-theoretical correlation loss is constructed. By combining the user-level consistency constraint and the cross-theoretical correlation loss, an alignment objective function is constructed to align the Big Five personality traits and MBTI representations.

2. The method according to claim 1, characterized in that, The construction of user feature representation specifically includes: For each piece of text from the user within a preset time window, a text semantic vector is extracted using a text encoder; Average pooling is performed on all text semantic vectors of the user to obtain a user-level semantic representation; Extract the user's language style feature vector and behavioral statistical feature vector; The user-level semantic representation, language style feature vector, and behavioral statistical feature vector are concatenated to form the user feature representation.

3. The method according to claim 1, characterized in that, The alignment objective function The formula is as follows: ； in, For cross-theoretical correlation loss, For user-level consistency constraints, These are the weighting coefficients.

4. The method according to claim 1, characterized in that, The generation of joint representations specifically includes: The aligned Big Five personality traits are input into a parameterized gating mapping network and processed by a nonlinear activation function to generate a gating weight vector consistent with the MBTI representation dimension. The gated weight vector is multiplied element-wise with the aligned MBTI representation to obtain the modulated MBTI representation. The aligned Big Five personality traits are added to the modulated MBTI traits to obtain the joint traits.

5. The method according to claim 1, characterized in that, Also includes: The joint representation is input into the downstream task network, and the task loss is constructed based on the supervision signal of the downstream task network. The representation fusion model is trained end-to-end using a joint loss function, which is obtained by weighted combination of the alignment objective function and the task loss. The representation fusion model includes the Big Five Personality Encoder, the MBTI Encoder, the alignment module, the representation fusion module, and the downstream task network.

6. The method according to claim 5, characterized in that, The joint loss function The formula is as follows: ； in, To align the balance parameters, Let the personality alignment loss function be... This is the loss function for downstream tasks.

7. A computer system comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method according to any one of claims 1-6.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of computer programs, which are used to be invoked by a processor and to execute the method as described in any one of claims 1-7.

9. A computer program product, comprising a computer program, characterized in that, The computer program is executed by a processor to implement the method according to any one of claims 1 to 6.

Images