Consumption willingness prediction method and device based on dual-path psychological mechanism and storage medium
By constructing a consumer intention prediction model based on a dual-path psychological mechanism, and combining static consumer traits with joint product characteristics, the problems of low prediction accuracy and poor robustness in existing technologies are solved, achieving high-precision and personalized consumer intention prediction.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- NANJING UNIV OF AERONAUTICS & ASTRONAUTICS
- Filing Date
- 2026-03-27
- Publication Date
- 2026-04-28
AI Technical Summary
Existing technologies lack a deep integration of consumer psychological characteristics in brand co-marketing, resulting in insufficient feature representation capabilities, low prediction accuracy, poor robustness, and an inability to achieve personalized predictions.
A consumption intention prediction method based on a dual-path psychological mechanism is adopted. By acquiring the static characteristics of consumers and the core functional characteristics and contextual association characteristics of the joint products, a dual-path psychological mechanism model is constructed, which is mapped to the intrinsic utility valuation and the extrinsic stimulus valuation respectively. The static characteristics of users are introduced to moderate the weight, so as to achieve high-precision and personalized prediction.
It significantly improves the interpretability and accuracy of the prediction model, enhances its adaptability to different groups with different traits, improves the robustness and generalization performance of the model, and achieves accurate consumer decision prediction.
Smart Images
Figure CN121937167A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of consumer intention quantification technology, and particularly relates to a method, device and storage medium for predicting consumer intention based on a dual-path psychological mechanism. Background Technology
[0002] In joint marketing practices, accurately predicting consumers' acceptance and purchase intentions for different co-branded products is crucial for companies to formulate marketing strategies. Current technologies for predicting consumer intentions often employ regression models based on historical sales data or classification models based on simple user profiles. These models lack consideration for the psychological characteristics of different users, leading to two major technical limitations: First, the model features are too simplistic, resulting in insufficient feature representation and a ceiling on prediction accuracy. Second, the models lack interpretability and adaptability; the same marketing strategy can exhibit significant prediction biases for different user groups, demonstrating poor robustness. Essentially, this is a technical challenge: how to construct a highly representative, interpretable, and heterogeneous adaptive predictive model.
[0003] The Elaboration Likelihood Model (ELM) provides a classic theoretical framework for deconstructing the dual pathways of attitude formation (the central pathway of deep cognition and the peripheral pathway of heuristic emotion), and has been widely applied in advertising and persuasion. While the dual-path concept of ELM is inspiring, existing techniques for applying it to brand association prediction have significant shortcomings: The mapping relationship is vague, failing to clearly and systematically map and operationalize the two core strategic dimensions of brand association (core structured features and contextual relevance) with the central and peripheral cues in ELM theory; the mechanism testing is one-sided, with most studies focusing only on the mediating role of a single path, lacking simultaneous empirical testing and comparison of two parallel psychological mechanisms within the same model framework; individual differences are ignored, especially the lack of in-depth exploration and quantitative modeling of how stable individual consumer traits can serve as systematic boundary conditions to differentiate the strength of the two paths. This prevents predictive models from achieving accurate, individualized predictions.
[0004] Therefore, there is an urgent need in this field for an integrated predictive method that can deeply integrate macro-strategic attributes, micro-psychological mechanisms, and individual differences. Summary of the Invention
[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method, device and storage medium for predicting consumer intentions based on a dual-path psychological mechanism. By simulating the inherent dual-path psychological mechanism of consumers' consumption decisions, the dynamic psychological characteristics of users are integrated into the construction and parameter tuning of a parameterized model, thereby achieving differentiated and high-precision prediction of the purchase intentions of consumer groups with different characteristics.
[0006] To achieve the above objectives, the present invention is implemented using the following technical solution:
[0007] In a first aspect, the present invention provides a method for predicting consumption intentions based on a dual-path psychological mechanism, comprising:
[0008] Obtain the core functional features and context-related features of the target joint product;
[0009] Obtain the static trait features of the target user; wherein, the static trait features include traditional traits and open traits;
[0010] The core functional features, context-related features, and static trait features are input into a pre-trained consumption intention assessment model to obtain a predicted consumption intention value. The consumption intention assessment model has two feature mapping branches, which map the core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates, respectively. One feature mapping branch uses traditional traits to moderate the mapping relationship between core functional features and intrinsic utility estimates, while the other feature mapping branch uses open-ended traits to moderate the mapping relationship between context-related features and extrinsic stimulus estimates.
[0011] Optionally, the core functional features of the target joint product are obtained through users' cognitive evaluation of the attributes of the target joint product itself;
[0012] The inherent attributes of the target joint products should include at least the product's appearance, taste, complimentary items, cost-effectiveness, raw material quality, health benefits, superior interactive design, and technological integration.
[0013] Optionally, the contextual association features of the target joint product are obtained through the user's perception of the marketing attributes of the target joint product;
[0014] The marketing attributes of a target joint product should include at least emotional response, cultural connotation, innovation, social attributes, unique experience, brand equity strategy, and attraction to new customer groups.
[0015] Optionally, the traditional traits are assessed using an introversion test; wherein the introversion test includes at least one of the following: whether one frequently thinks about how to avoid failure in life, whether one frequently imagines experiencing bad things that one fears will happen, and whether one frequently worries about what kind of person one will become in the future;
[0016] The openness trait is assessed using an extraversion test; the extraversion test includes at least one of the following: whether one frequently imagines how one will achieve one's ideals and aspirations, whether one always focuses on the future achievements one hopes to attain, and whether one frequently thinks about what kind of person one ideally wants to become in the future.
[0017] Optionally, the consumption intention assessment model includes a weighted aggregation layer, and a first mapping sub-model and a second mapping sub-model connected in parallel to the input of the weighted aggregation layer;
[0018] The first and second mapping sub-models are linear regressors or shallow neural networks with adjustable mapping coefficients;
[0019] The weighted aggregation layer has an embedded nonlinear function with adjustable weight coefficients.
[0020] Optionally, the training method for the consumption intention assessment model includes:
[0021] Obtain a training sample set; where each training sample is a tuple, including the core functional features and contextual association features of historical joint products, and the static trait features of the corresponding historical users;
[0022] The first mapping sub-model maps the core functional features in the input training samples to intrinsic utility estimates; wherein, the mapping weights of the first mapping sub-model are modulated by the traditional traits in the input training samples;
[0023] The context-related features in the input training samples are mapped to external stimulus estimates through a second mapping sub-model; wherein, the mapping weights of the second mapping sub-model are modulated by the openness traits in the input training samples;
[0024] The intrinsic utility valuation and extrinsic stimulus valuation are weighted and fused by a weighted aggregation layer to output a predicted value of consumer willingness.
[0025] Loss is calculated based on the intrinsic utility estimate and the actual intrinsic utility value pre-labeled with core functional features, the extrinsic stimulus estimate and the actual extrinsic stimulus value pre-labeled with contextual association features, and the consumption intention prediction and the actual consumption intention value pre-labeled with the input training sample. Among them, the actual intrinsic utility value is obtained by the target user's evaluation of the value characteristics of the target joint product; the actual extrinsic stimulus value is obtained by the target user's evaluation of the novelty of the target joint product.
[0026] Based on the loss calculation results, the mapping weights of the first and second mapping sub-models, the adjustment strength of the traditional trait on the mapping weights of the first mapping sub-model, and the adjustment strength of the open trait on the mapping weights of the second mapping sub-model are adjusted to obtain a well-trained consumer willingness assessment model.
[0027] Optionally, the formula for calculating the intrinsic utility estimate includes:
[0028] ,
[0029] in, Valuing intrinsic utility As the core functional feature, Traditional characteristics The baseline mapping weights for the first mapping sub-model. This represents the adjustment strength of the mapping weights of the traditional trait on the first mapping sub-model.
[0030] Optionally, the formula for calculating the external stimulus valuation includes:
[0031] ,
[0032] in, Valuation based on external stimuli For context-related features, It is an open-ended trait. The baseline mapping weights for the second mapping sub-model. The moderating strength of the mapping weights of the second mapping sub-model for open-type traits.
[0033] Secondly, the present invention provides a consumer intention prediction device based on a dual-path psychological mechanism, comprising:
[0034] Marketing Product Feature Acquisition Module: Used to acquire the core functional features and contextual association features of the target joint product;
[0035] User Feature Acquisition Module: Used to acquire static trait features of target users; wherein, the static trait features include traditional traits and open traits;
[0036] The consumption intention assessment module is used to input the core functional features, context-related features, and static trait features into a pre-trained consumption intention assessment model to obtain a predicted consumption intention value. The consumption intention assessment model has two feature mapping branches, which map the core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates, respectively. One feature mapping branch uses traditional traits to moderate the mapping relationship between core functional features and intrinsic utility estimates, while the other feature mapping branch uses open-ended traits to moderate the mapping relationship between context-related features and extrinsic stimulus estimates.
[0037] Thirdly, the present invention provides a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the consumption intention prediction method based on a dual-path psychological mechanism as described in any of the first aspects.
[0038] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: By constructing a consumer intention prediction model that integrates dual-path psychological mechanisms, the core functional features and contextual association features of brand association are mapped to the central path (intrinsic utility valuation) and the peripheral path (extrinsic stimulus valuation), respectively. This effectively solves the technical defects of existing models, such as single feature representation and failure to distinguish heterogeneous psychological mechanisms, significantly improving the interpretability and prediction accuracy of the model. Furthermore, by introducing static user traits (traditional traits and open traits) as moderating variables, the mapping weights of the two paths are dynamically adjusted, enabling the model to adapt to user heterogeneity. This overcomes the problem of large prediction bias for different trait groups in traditional one-size-fits-all prediction strategies, enhancing the robustness and generalization performance of the model. Moreover, by adopting a multi-task learning framework and simultaneously optimizing the main prediction task and auxiliary prediction task through a comprehensive loss function, the stability and convergence efficiency of model training are further improved. This achieves refined simulation and personalized prediction of the consumer decision-making process, providing quantifiable and interpretable technical support for precise marketing decisions by enterprises. Attached Figure Description
[0039] Figure 1 The diagram shown is a flowchart of a consumption intention prediction method based on a dual-path psychological mechanism in one embodiment of the present invention.
[0040] Figure 2 The diagram shown is an architecture diagram of the dual-path psychological feature fusion module in one embodiment of the present invention;
[0041] Figure 3 The diagram shown illustrates the interaction effect of core functional features and traditional characteristics on intrinsic utility in one embodiment of the present invention.
[0042] Figure 4 The diagram shown illustrates the interaction effect of context-related features and openness traits on external stimuli in one embodiment of the present invention.
[0043] Figure 5 The diagram shown is a flowchart of the training process for a consumer intention prediction model in one embodiment of the present invention. Detailed Implementation
[0044] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.
[0045] Example 1
[0046] like Figure 1 As shown, this embodiment provides a method for predicting consumption intentions based on a dual-path psychological mechanism, specifically including the following steps:
[0047] Step S101: Obtain the core functional features and contextual characteristics of the target joint product, as well as the static trait characteristics of the target user.
[0048] Enterprise customers initiate prediction requests, which include the core functional features and contextual characteristics of the target joint product, as well as a list of target user groups.
[0049] The core functional characteristics of the target co-branded product are obtained through users' cognitive evaluation of the product's own attributes. In this embodiment, the attributes of the target co-branded product include at least product appearance, taste, free gifts, cost-effectiveness, raw material quality, health benefits, superior interactive design, and technological integration. A five-point Likert scale is used for measurement, with 1 representing "strongly disagree," 2 representing "disagree," 3 representing "neutral," 4 representing "agree," and 5 representing "strongly agree." Specifically, the following three items are used for measurement: "The two brands complement each other functionally," "This co-branded product has a significant functional upgrade," and "This co-branding combines the functional advantages of both brands." The average score is taken as the score for the core functional characteristics.
[0050] The contextual relevance characteristics of the target co-branded product are obtained through users' perception of its marketing attributes. These attributes include at least emotional response, cultural connotation, innovation, social attributes, unique experience, brand equity strategy, and appeal to new customer groups. A five-point Likert scale is used, with three items—"The images of these two brands match well," "The visual design of this collaboration is harmonious," and "The two brands look very coordinated together"—for measurement. The average score is then used as the score for the contextual relevance characteristics.
[0051] The static trait characteristics include traditional and open-ended traits. Based on the Consumer Innovative-Conservative Scale, the user's tendencies in both traditional and open-ended dimensions are measured. Each dimension also consists of three questions using a Likert five-point scale. The traditional trait is measured using three items: "I often think about how to avoid failure in life," "I often imagine myself experiencing bad things that I fear," and "I often worry about what kind of person I will become in the future." The open-ended trait is measured using three items: "I often imagine how I will achieve my ideals and ambitions," "I always focus on the future achievements I hope to attain," and "I often think about what kind of person I ideally want to become in the future." The average of the item results is taken as the user's trait score, which is recorded as the traditional trait and the open-ended trait, respectively.
[0052] All raw data was collected using a structured questionnaire and standardized using Z-scores before being stored in the feature database. Enterprise clients push questionnaires containing the above scales to their target users by calling the system. After users complete the responses, the data is sent back to the server. First-time users also need to complete a feature questionnaire, and the system vectorizes the results and stores them in the user profile database.
[0053] Step S102: Input the core functional features and context-related features into the trained consumer intention assessment model to obtain the predicted consumer intention value.
[0054] like Figure 2 As shown, the consumption intention assessment model has two feature mapping branches, which respectively map core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates. (Reference) Figure 3 One of the feature mapping branches uses traditional trait modulators to adjust the mapping relationship between core functional features and intrinsic utility estimates, referencing... Figure 4 Another feature mapping branch uses open-ended traits to moderate the mapping relationship between context-related features and external stimulus estimates. The two feature mapping branches constitute a dual-path psychological feature fusion module, which is implemented on the server as a machine learning model, containing two linear sub-models with adjustable coefficients, corresponding to the central path and peripheral path in ELM theory, respectively.
[0055] The consumption intention assessment model includes a first mapping sub-model and a second mapping sub-model in parallel, and a weighted aggregation layer connected in series to the outputs of the first mapping sub-model and the second mapping sub-model, respectively. The first mapping sub-model and the second mapping sub-model are linear regressors or shallow neural networks with adjustable mapping coefficients, and the weighted aggregation layer has an embedded nonlinear function with adjustable weight coefficients.
[0056] Calculation of the first mapping sub-model (central path):
[0057] Let the core functional features be Traditional characteristics are Then the intrinsic utility valuation The calculation formula is:
[0058] ,
[0059] in, The baseline mapping weights for the first mapping sub-model. This represents the adjustment strength of the mapping weights of the traditional trait on the first mapping sub-model.
[0060] Calculation of the second mapping sub-model (edge path):
[0061] Let the context-related features be: Openness is a characteristic External stimuli to valuation The calculation formula is:
[0062] ,
[0063] in, The baseline mapping weights for the second mapping sub-model. The moderating effect of openness traits on the mapping weights of the second mapping sub-model.
[0064] Calculation of weighted aggregation layer:
[0065] The intention forecasting module integrates intrinsic utility estimates and extrinsic stimulus estimates into a final consumption intention forecast. This embodiment uses a trainable weighted summation method:
[0066] ,
[0067] in, and These are trainable parameters, initially set to 0.5, allowing the model to automatically adjust the relative importance of the two paths based on the data.
[0068] The prediction results are returned, usually in JSON format, including each user's predicted intention score, average intention score, and intention distribution of different trait groups, and can be synchronized to the customer's backend for visualization in chart form.
[0069] like Figure 5 As shown, in this embodiment, the training method of the consumption intention assessment model includes the following steps:
[0070] Step S201: Obtain the training sample set
[0071] Each training sample is a plural group, including the core functional features and contextual association features of historical joint products, as well as the static trait features of the corresponding historical users. To obtain the labeled data required for supervised learning, the system uses an online A / B testing framework as the training data generator.
[0072] Experimental Design: The system uses different co-branded products (e.g., two brand co-branded schemes A and B) as experimental variables. Each user only interacts with one co-branded product to avoid cross-interference.
[0073] User grouping and sampling control: To ensure the comparability of the experimental group and the control group, the system adopts a stratified random sampling method. First, based on the traditional characteristics in the user profile, the sampling is performed. With openness Users are divided into high and low score groups, and then randomly assigned to different joint product groups within each tier to ensure a balanced distribution across key trait dimensions. Simultaneously, the same user is not repeatedly involved in the same type of experiment within a short period to avoid the impact of learning or fatigue effects on label quality.
[0074] Tag Data Collection: After users come into contact with the joint product, the system immediately pushes a tag collection questionnaire to measure three types of tags. Each question uses the same 5-point scale as the feature measurement to ensure scale consistency.
[0075] Intrinsic utility real value The measurement was conducted using four items: "I think this co-branded product is useful to me," "Using this co-branded product can meet my needs," "This co-branded product can effectively help me achieve a certain goal," and "This co-branded product has high practical value."
[0076] External stimulus true value The measurement uses four criteria: "This product is innovative," "This product is novel and unique," "This product is highly original," and "This product is creative." This effectively captures consumers' subjective evaluations of a product's novelty and originality.
[0077] Real value of consumer willingness The measurement was conducted using three items: "I am very likely to purchase this product," "I am very likely to consider purchasing this product," and "I am very willing to purchase this product."
[0078] Data alignment and storage: Integrating user, joint product, and feature data ( ) and tag data ( After aligning the data and removing missing or obviously incomplete data, complete training samples are formed and stored in the training database for use during model training.
[0079] Step S202: Model forward propagation
[0080] The first mapping sub-model maps the core functional features in the input training samples to intrinsic utility estimates. The mapping weights of the first mapping sub-model are influenced by the traditional features in the input training samples. The adjustment.
[0081] The second mapping sub-model maps the contextual features in the input training samples to external stimulus estimates. The mapping weights of the second mapping sub-model are influenced by the openness characteristics in the input training samples. The adjustment.
[0082] The intrinsic utility valuation and extrinsic stimulus valuation are weighted and fused through a weighted aggregation layer to output a predicted value of consumer willingness. .
[0083] Step S203: Loss Calculation and Backpropagation
[0084] Loss is calculated based on the intrinsic utility estimate and the actual intrinsic utility value pre-labeled with core functional features, the extrinsic stimulus estimate and the actual extrinsic stimulus value pre-labeled with contextual association features, and the consumption intention prediction and the actual consumption intention value pre-labeled with the input training sample.
[0085] Using a comprehensive loss function Simultaneously considering the prediction errors of the main prediction task (willingness to consume) and two auxiliary prediction tasks (intrinsic utility and extrinsic stimuli):
[0086] ,
[0087] in, Let the mean square error function be used. and is the regularization coefficient, and is a hyperparameter greater than 0, used to balance the weights of the main task and auxiliary tasks. In this embodiment, optimization is performed through a validation set. and All are set to 0.3.
[0088] Step S204: Parameter Optimization
[0089] Mini-batch gradient descent was used for parameter optimization. The training dataset was divided into batches of size 64, and the loss was calculated and gradients were backpropagated for each batch. The optimizer Adam was selected, with an initial learning rate of 0.001 that decayed exponentially with each training epoch.
[0090] During training, after each complete data traversal (epoch), the loss is calculated on the validation set. When the validation loss no longer decreases for 10 consecutive epochs, an early stopping mechanism is triggered to prevent overfitting.
[0091] Step S205: Model Evaluation and Deployment
[0092] Load the sample set from the training database and divide it into a training set (70%), a validation set (15%), and a test set (15%). Initialize the model parameters. It is a random small value.
[0093] After training, evaluate the prediction accuracy of the final model on the test set (including root mean square error RMSE, coefficient of determination). The model parameters that meet the accuracy requirements are then fixed to obtain the trained consumer intention prediction model. The trained model is then packaged into an application service.
[0094] Example 2
[0095] This embodiment provides a consumer intention prediction device based on a dual-path psychological mechanism, including:
[0096] Marketing Product Feature Acquisition Module: Used to acquire the core functional features and contextual association features of the target joint product;
[0097] User Feature Acquisition Module: Used to acquire the static trait features of the target user, wherein the static trait features include traditional traits and open traits;
[0098] The consumption intention assessment module is used to input the core functional features, context-related features, and static trait features into a pre-trained consumption intention assessment model to obtain predicted consumption intention values. The consumption intention assessment model has two feature mapping branches, which map the core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates, respectively. One feature mapping branch uses traditional trait modifiers to moderate the mapping relationship between core functional features and intrinsic utility estimates, while the other feature mapping branch uses open-ended trait modifiers to moderate the mapping relationship between context-related features and extrinsic stimulus estimates.
[0099] The device provided in this embodiment can execute the consumption intention prediction method based on the dual-path psychological mechanism provided in any step of Embodiment 1, and has the corresponding functional modules and beneficial effects of the execution method.
[0100] Example 3
[0101] This embodiment provides a computer storage medium on which a computer program is stored. When the computer program is executed by a processor, it implements the consumption intention prediction method based on the dual-path psychological mechanism provided in any step of Embodiment 1.
[0102] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0103] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0104] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0105] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0106] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A method for predicting consumption intentions based on a dual-path psychological mechanism, characterized in that, include: Obtain the core functional features and context-related features of the target joint product; Obtain the static trait features of the target user; wherein, the static trait features include traditional traits and open traits; The core functional features, context-related features, and static trait features are input into a pre-trained consumption intention assessment model to obtain a predicted consumption intention value. The consumption intention assessment model has two feature mapping branches, which map the core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates, respectively. One feature mapping branch uses traditional traits to moderate the mapping relationship between core functional features and intrinsic utility estimates, while the other feature mapping branch uses open-ended traits to moderate the mapping relationship between context-related features and extrinsic stimulus estimates.
2. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 1, characterized in that, The core functional features of the target joint product are obtained through users' cognitive evaluation of the target joint product's own attributes; The inherent attributes of the target joint products should include at least the product's appearance, taste, complimentary items, cost-effectiveness, raw material quality, health benefits, superior interactive design, and technological integration.
3. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 1, characterized in that, The contextual relevance of the target joint product is obtained through users' perception of the marketing attributes of the target joint product; The marketing attributes of a target joint product should include at least emotional response, cultural connotation, innovation, social attributes, unique experience, brand equity strategy, and attraction to new customer groups.
4. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 1, characterized in that, The traditional traits were assessed using introversion tests. The openness trait was assessed using an extraversion test.
5. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 1, characterized in that, The consumption intention assessment model includes a weighted aggregation layer, and a first mapping sub-model and a second mapping sub-model connected in parallel to the input of the weighted aggregation layer; The first and second mapping sub-models are linear regressors or shallow neural networks with adjustable mapping coefficients; The weighted aggregation layer has an embedded nonlinear function with adjustable weight coefficients.
6. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 5, characterized in that, The training method for the consumption intention assessment model includes: Obtain a training sample set; where each training sample is a tuple, including the core functional features and contextual association features of historical joint products, and the static trait features of the corresponding historical users; The first mapping sub-model maps the core functional features in the input training samples to intrinsic utility estimates; wherein, the mapping weights of the first mapping sub-model are modulated by the traditional traits in the input training samples; The context-related features in the input training samples are mapped to external stimulus estimates through a second mapping sub-model; wherein, the mapping weights of the second mapping sub-model are modulated by the openness traits in the input training samples. The intrinsic utility valuation and extrinsic stimulus valuation are weighted and fused by a weighted aggregation layer to output a predicted value of consumer willingness. Loss is calculated based on the intrinsic utility estimate and the actual intrinsic utility value pre-labeled with core functional features, the extrinsic stimulus estimate and the actual extrinsic stimulus value pre-labeled with contextual association features, and the consumption intention prediction and the actual consumption intention value pre-labeled with the input training sample. Among them, the actual intrinsic utility value is obtained by the target user's evaluation of the value characteristics of the target joint product; the actual extrinsic stimulus value is obtained by the target user's evaluation of the novelty of the target joint product. Based on the loss calculation results, the mapping weights of the first and second mapping sub-models, the adjustment strength of the traditional trait on the mapping weights of the first mapping sub-model, and the adjustment strength of the open trait on the mapping weights of the second mapping sub-model are adjusted to obtain a well-trained consumer willingness assessment model.
7. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 6, characterized in that, The formula for calculating the intrinsic utility valuation includes: ， in, Valuing intrinsic utility As the core functional feature, Traditional characteristics The baseline mapping weights for the first mapping sub-model. The adjustment strength of the traditional trait on the mapping weight of the first mapping sub-model.
8. The consumption intention prediction method based on a dual-path psychological mechanism according to claim 6, characterized in that, The formula for calculating the valuation of external stimuli includes: ， in, Valuation based on external stimuli For context-related features, It is an open-ended trait. The baseline mapping weights for the second mapping sub-model. The moderating strength of the mapping weights of the second mapping sub-model for open-type traits.
9. A device for predicting consumption intentions based on a dual-path psychological mechanism, characterized in that, include: Marketing Product Feature Acquisition Module: Used to acquire the core functional features and contextual association features of the target joint product; User Feature Acquisition Module: Used to acquire static trait features of target users; wherein, the static trait features include traditional traits and open traits; The consumption intention assessment module is used to input the core functional features, context-related features, and static trait features into a pre-trained consumption intention assessment model to obtain a predicted consumption intention value. The consumption intention assessment model has two feature mapping branches, which map the core functional features and context-related features to intrinsic utility estimates and extrinsic stimulus estimates, respectively. One feature mapping branch uses traditional traits to moderate the mapping relationship between core functional features and intrinsic utility estimates, while the other feature mapping branch uses open-ended traits to moderate the mapping relationship between context-related features and extrinsic stimulus estimates.
10. A computer storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the consumption intention prediction method based on the dual-path psychological mechanism as described in any one of claims 1-8.