A dynamic adaptation method for creative expression based on a multicultural framework
By building a multicultural framework and dynamic identification mechanism, the adaptation problem of the content recommendation system in a multicultural context is solved, the precise matching and cultural adaptability of creative content are achieved, and creative expression and communication in a multicultural environment are promoted.
Patent Information
- Application Number
- CN202311690388.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-12-11
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-12-11
AI Technical Summary
Existing content recommendation systems are difficult to accurately identify and adapt to the needs of different cultural groups in a multicultural context, resulting in poor creative dissemination, and large language models face the challenge of cultural complexity recognition in personalized content recommendations.
Build a multicultural framework, acquire cultural characteristics through social media, user surveys and online behavior records, establish a cultural distance matrix, design dynamic recognition mechanisms and cross-cultural language adaptation models, generate creative content that takes into account innovation and diversity, and combines user feedback to optimization.
It achieves accurate matching between creative content and the receiving group, improves cultural adaptability and participation, and promotes creative expression and communication in a multicultural environment.
Smart Images

Figure CN117763228B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of information technology, and in particular to a creative expression dynamic adaptation method based on a multicultural framework. Background Art
[0002] With the acceleration of globalization and the development of digital technology, the world is becoming increasingly connected. This global interconnection not only promotes economic and social integration, but also brings opportunities and challenges for cultural exchanges. People from different cultural backgrounds meet on the Internet to share and exchange their views and ideas. This cultural diversity provides a rich soil for creative expression, but it also brings the problem of how to effectively communicate and spread creativity between different cultures. In this context, content creators and recommendation systems face a major challenge, which is how to understand and adapt to the needs and preferences of different cultural groups. Most existing content recommendation systems rely on generalized algorithms that often ignore the impact of cultural diversity. This approach may be effective in some cultural contexts, but may not be applicable in other cultural contexts, or even cause misunderstandings. In this context, large language models, such as the GPT series, have shown great potential in personalized content recommendation and creative generation. These models are able to process large amounts of text data, learn the language and expressions of different cultural contexts, and generate creative and rich content based on this. With a deep understanding of different cultural contexts, these models can help create more diverse and personalized content, thereby improving the creativity and engagement of different cultural groups. However, a major challenge facing these models is how to accurately identify and divide different cultural groups. Since culture is a multidimensional and dynamically changing concept, a single or static group classification often fails to accurately capture the complexity of culture. Therefore, developing a creative expression method that can adapt to a multicultural environment is an urgent technical need at this stage. Summary of the invention
[0003] The present invention provides a creative expression dynamic adaptation method based on a multicultural framework, which mainly includes:
[0004] Construct a multicultural framework that includes mainstream values, cognitive patterns, and aesthetic preference dimensions, and create a cultural distance matrix to identify the heterogeneity between cultural groups; design a dynamic identification mechanism for cultural groups based on the differences in the multicultural framework to evaluate the cultural distance between different groups in real time; construct a cross-cultural language adaptation model to support the dynamic matching and adaptation between creative content expression and the receiving group; generate a candidate creative content set that takes into account both innovation and diversity based on a generative adversarial network, covering cross-cultural expressions; evaluate the distance between cultural groups and screen matching creative content based on the user's cultural characteristics and interaction data; design a user simulation mechanism to predict the acceptance preferences of different cultural groups and correct the adaptation errors of creative content expression; collect user feedback through A / B testing to evaluate the acceptance effect of creative content in multicultural groups; adjust the cultural distance between groups in the multicultural framework based on the feedback effect to achieve dynamic optimization of expression adaptation; identify the taste drift trend of different groups based on user browsing history and feedback data, and optimize content recommendation strategies.
[0005] In one embodiment, the construction of a multicultural framework including mainstream values, cognitive patterns, and aesthetic preference dimensions, and the creation of a cultural distance matrix to identify heterogeneity between cultural groups includes:
[0006] Through social media, user surveys, and online behavior records, we obtain data on users' mainstream values, cognitive patterns, and aesthetic preferences, build a meta-cultural framework that includes multiple cultural characteristics and user behavior dimensions, and determine the importance and interaction methods of each dimension in the framework; extract the cultural characteristics of different cultural groups, including language habits, values, and artistic preferences, through the pages visited by users, their dwelling time, and click order; according to the cultural characteristics of different cultural groups, use the principal component analysis algorithm to train the model, build a cultural characteristic model, model the cultural characteristics of different cultural groups, and obtain vectors that represent the cultural characteristics of different cultural groups; use the method of calculating the Euclidean distance to calculate the distance between different cultural characteristic vectors. Based on the cultural distance matrix, a cultural distance matrix is constructed to determine the relative positions and differences between different cultural groups. Based on the cultural distance matrix, the K-means clustering algorithm is used to divide different users into different cultural groups and identify the heterogeneity between cultural groups. The online and offline data of users are obtained, and the cultural feature vectors of users are extracted through the established cultural feature model. The online data includes browsing history and search habits, and the offline data includes cultural activities participated in and geographic location information. Based on the distance between the user's cultural feature vector and the vectors of different cultural groups, the cultural group to which the user belongs is determined. By continuously tracking user data and monitoring the changes in the user's cultural feature vector, a dynamic evaluation of the user's cultural group affiliation is achieved.
[0007] In one embodiment, the design is based on a dynamic identification mechanism of cultural groups based on the differences in a multicultural framework, and the cultural distance between different groups is evaluated in real time, including:
[0008] Extract the cultural characteristics of each cultural group, including language habits, values, and artistic preferences, and construct a feature vector for each cultural group; if new user data flows in, extract the user's cultural characteristics through the pages visited by the user, the duration of stay, and the order of clicks, and construct a vector representing the user's cultural characteristics; compare the user's cultural feature vector with the vectors of each cultural group in the framework, and use the Euclidean distance method to calculate the distance between the user vector and the vectors of each cultural group; set a threshold based on the distance between the user vector and the vectors of each group, and use the threshold to determine the cultural group to which the user is most likely to belong; continuously obtain new user data, and repeat the process of feature extraction, vector calculation, and distance judgment; if the change in the distance between the user vector and the original belonging group vector exceeds the predetermined threshold, it is determined that the user's cultural group has changed, and the user's cultural group affiliation is adjusted; based on the update of all user cultural vectors, evaluate the changes in the members within each cultural group, and correct the cultural group feature vector.
[0009] In one embodiment, the construction of a cross-cultural language adaptation model to support dynamic matching and adaptation between creative content expression and receiving groups includes:
[0010] Obtain the cultural feature vector of the target user and determine the cultural group to which he belongs; calculate the cultural distance between the cultural group to which the target user belongs and other cultural groups based on the constructed cultural distance matrix; use the linear regression algorithm to train the model based on the cultural distance data, user feature vector and historical feedback data, build a cross-cultural language adaptation model, and predict the degree of adaptation of the target user's cultural group to various creative content expressions; according to the adaptation evaluation results, select the creative content expression with the highest degree of adaptation to the target user's cultural group, match and generate creative content of the corresponding method, and deliver it to the user; obtain user feedback on the recommended content after the user browses, clicks, and forwards; if the user feedback is positive, increase the weight of the corresponding cultural distance and the creative content expression; if the user feedback is negative, reduce the matching weight of the corresponding distance and the expression method, and determine the best matching option; optimize the cross-cultural language adaptation model by continuously adjusting the weights to achieve dynamic matching between the creative content expression method and the receiving group.
[0011] In one embodiment, the candidate creative content set based on the generative adversarial network that takes into account both innovation and diversity, covering cross-cultural expressions, includes:
[0012] Obtain various existing creative content as training samples, and the samples cover different creative expressions, including content from different cultural backgrounds; obtain labeled training samples, use the label information to enhance the samples, increase sample diversity through data expansion technology, and generate a diverse set of candidate creative content; based on the enhanced data set, build and train a generative adversarial network, which includes a generative model and a discriminative model. The generative model generates creative content, and the discriminative model evaluates the authenticity and quality of the content; use the trained generative model and input random noise conditions to generate candidate creative content; if the quality of the generated content is lower than expected, obtain user feedback on some candidate content to evaluate the innovation and diversity of the content; if the user feedback result is not good, further optimize the generative adversarial network based on the feedback data to improve the quality of the content generated by the model; repeat the generation, evaluation and optimization process until the generated creative content meets the preset innovation and diversity requirements.
[0013] In one embodiment, the step of evaluating the distance between cultural groups and screening matching creative content based on the cultural characteristics and interaction data of the user includes:
[0014] Obtain the user's cultural feature vector, determine the cultural group to which the target user belongs, and extract the feature vector of the cultural group; use the Euclidean distance method to calculate the cultural distance between the target user vector and the cultural group vector to which it belongs; according to the preset threshold, determine the cultural group whose cultural distance with the target user is less than the preset threshold in the constructed cultural distance matrix; based on the generated creative content set, extract the semantic feature vector of each creative content, and the semantic features include theme features, emotional features, and style features; calculate the similarity between each creative content feature vector and the target user group feature vector; if the similarity is higher than the preset threshold, add the corresponding creative content to the candidate set; set the reward value of positive feedback according to the feedback data of user browsing, liking, and forwarding, and if the user feedback level is greater than the preset threshold, increase the matching weight of the corresponding creative content and the cultural distance; adjust the matching weight of the creative content and the cultural distance through the Q-learning algorithm to optimize the recommendation strategy; screen out the creative content that matches the cultural distance of the target group and obtains positive feedback from users to form a recommendation set; also includes: designing and optimizing the dynamic scoring mechanism for cross-cultural content recommendation based on the analysis of cultural features and user behavior data.
[0015] The dynamic scoring mechanism for cross-cultural content recommendation is designed and optimized based on the analysis of cultural characteristics and user behavior data, including:
[0016] According to cultural characteristics and user behavior patterns, determine the impact of different cultural characteristics on user preferences for recommended content, and construct a comprehensive scoring formula C = α*(F_u-F_c)+β*D+γ*T, where F_u represents the user's cultural feature vector, F_c represents the average feature vector of the cultural group to which the user belongs, D represents the user's behavior data, and T represents the time series data of the user's interaction with the content. α, β, γ are initial weight coefficients set based on theoretical research and adjusted according to actual data. Use API to obtain user cultural feature data, behavior data, and time series data. Perform preprocessing operations such as cleaning and normalization on the obtained data. Use the obtained data to train a recursive neural network model and predict the cultural adaptation score C based on real-time data. Design and implement a real-time user feedback mechanism to obtain user feedback on recommended content. According to the obtained user feedback, adjust the weight coefficients α, β, γ and optimize the scoring formula. According to the optimized scoring formula, dynamically adjust the logic of the recommendation system. Continuously monitor changes in user preferences and the effectiveness of the recommendation system. Implement a continuous learning mechanism for the model to adapt to changes in user behavior and cultural characteristics. Regularly evaluate model effectiveness and continuously adjust the model based on feedback and analysis results.
[0017] In one embodiment, the design of the user simulation mechanism predicts the acceptance preferences of different cultural groups and corrects the adaptation error of the creative content expression method, including:
[0018] Obtain user data of different cultural groups, including browsing and clicking behavior data; extract vectors representing user characteristics, including points of interest and activity participation; recommend creative content with different expressions to each cultural group, and record user feedback data on the creative content, including browsing and clicking behavior data; use a decision tree algorithm to establish a mapping model between group user feature vectors and content expression methods based on the obtained user feedback data, and predict the feedback of users of a certain cultural group on content with a given expression method; in the recommendation process, compare the expression method of the creative content with the user feedback predicted by the model. If there is an adaptation error between the expression method of a certain creative content and the user feedback predicted by the model, then select the content expression method with the most positive feedback in the simulation results; use a collaborative filtering algorithm to train the model based on the user's own characteristics, and generate personalized expression preference predictions; continuously obtain user feedback and adjust the personalized model to continuously optimize the adaptation of creative content, gradually approach the user's expression preference, and realize creative content recommendation; also includes: based on cultural characteristics and user behavior analysis, realize dynamic prediction and adaptation of user preferences in a multicultural environment.
[0019] The dynamic prediction and adaptation of user preferences in a multicultural environment based on cultural characteristics and user behavior analysis specifically includes:
[0020] According to the behaviors and preferences of users from different cultural backgrounds, identify the key cultural features that affect users’ content preferences, including language habits, aesthetic standards, and values, build a framework that includes multiple cultural features and user behavior dimensions, determine the importance and interaction mode of each dimension in the framework, and design a preliminary prediction formula S = ∑(w_i*C_i) + δ(B). Among them, S represents the cultural adaptability score of users to creative content, C_i represents the i-th cultural feature, w_i is the weight of the corresponding cultural feature, B represents user behavior data, and δ is the function that processes user behavior data. Obtain relevant data of users from different cultural backgrounds, including the behaviors and feedback of users from different cultural backgrounds. Apply the prediction formula based on the obtained data to test its effectiveness and accuracy. According to the test results, adjust the definition of the cultural feature weight w_i and the function δ to optimize the formula. Continuously monitor the performance of the formula in actual applications and obtain real-time feedback from users. According to the new data and feedback obtained, continuously iterate and update the formula to maintain the timeliness and accuracy of the formula.
[0021] In one embodiment, collecting user feedback through A / B testing to evaluate the acceptance effect of creative content in a multicultural group includes:
[0022] According to the cultural characteristics of users, users are divided into different cultural groups to construct a multicultural user group data set; samples are randomly selected from the candidate creative content set, and the samples randomly selected from the candidate creative content set are divided into two mutually exclusive test groups, Group A and Group B, according to the proportion of the cultural groups to which the users belong; the traffic of the website or application is redirected to Group A and Group B, and the users who visit the content of Group A and Group B represent the corresponding cultural groups respectively; after the users browse the content, the user's behavior data is obtained, including the page dwell time, click-through rate, and conversion rate; the acceptance effect of Group A / B is calculated based on the user's behavior data; the acceptance effect of Group A / B is compared through t-test to determine whether the difference in the acceptance effect of the two test groups exceeds the preset threshold; if the test results show that the difference in the acceptance effect of different cultural groups on creative content exceeds the preset threshold, it proves that there is a difference in the acceptance effect of different cultural groups on creative content, and the content composition strategy is adjusted according to the acceptance effect of different cultural groups on creative content; by conducting A / B tests regularly, the acceptance effect of each cultural group on creative content is continuously monitored and evaluated, and the creative content generation strategy is continuously revised and optimized based on the test results.
[0023] In one embodiment, the method of adjusting the cultural distance between groups in a multicultural framework based on feedback effects to achieve dynamic optimization of expression adaptation includes:
[0024] Obtain the behavior data of users browsing, clicking and forwarding the recommended creative content; use the K-means algorithm to train the model based on the content style and conversion rate indicators to determine the differences in the feedback of users from different cultural groups on the expression of creative content; recalculate the cultural distance between cultural group vectors based on the full user behavior data using the method of calculating the Euclidean distance. The full user behavior data refers to the behavior data generated by all users in the process of using the platform or service, including the user's browsing history, click behavior, content preference, forwarding and social interaction data; after the vector distance is updated, re-evaluate the mapping relationship between users and cultural groups; based on the new cultural distance, re-evaluate the acceptance of content with different expressions in each cultural group, and form a new matching table between content expression and cultural distance; if the matching weights of certain distances and content expression need to be adjusted, the matching table is updated accordingly; re-recommend content and verify the conversion rate of users under the new method. If the conversion rate of users under the new method is improved, it is considered that the optimization is effective; through closed-loop iterative optimization, obtain a dynamic decision table matching cultural distance and expression method, and realize continuous upgrading of expression adaptation.
[0025] In one embodiment, identifying the taste drift trends of different groups based on user browsing history and feedback data, and optimizing the content recommendation strategy, includes:
[0026] Continuously obtain user feature data, which includes browsing history, click-through rate, and feedback, to detect changes in user personal tastes; use the ARIMA algorithm to train models based on user feature data, build a taste drift model, and detect fluctuations in user tastes; if it is detected that user tastes fluctuate, mark the user sets with positive and negative taste migrations; based on users with changing tastes, explore the distribution patterns of users with different taste migration directions in terms of age and gender; based on the collective drift trend of the tastes of user groups, determine the changing direction of different cultural groups' acceptance of creative content; based on the user's cultural feature vector, screen creative content that matches the user's cultural preferences, and adjust the expression of the selected content, including adjusting the language style, visual presentation, or theme framework of the content; for users with personalized taste preferences, establish an exclusive personalized taste model, and in the recommendation process, form a personalized recommendation plan based on the taste of the user's group and the user's own personalized taste; after the user browses the creative content, obtain user feedback data to detect the effectiveness of the taste drift model; if the effect deviates, adjust the model parameters based on the feedback data to achieve closed-loop iterative optimization.
[0027] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:
[0028] The present invention proposes a dynamic adaptation method for creative expression based on a multicultural framework. The method accurately captures and displays the uniqueness and differences between different cultural groups by integrating dimensions such as mainstream values, cognitive patterns, and aesthetic preferences. By evaluating cultural distance in real time, the present invention can identify creative content suitable for different cultural groups and ensure the matching degree between the content and the receiving group.
[0029] The present invention uses cross-cultural language adaptation strategies to ensure that creative content can be consistent with the language and expression of the target cultural group. It can also generate content that takes into account both innovation and diversity to meet the needs of users from different cultural backgrounds. Through user simulation mechanisms and feedback collection, the present invention can accurately predict the acceptance preferences of different cultural groups and optimize content expression accordingly.
[0030] The present invention improves the cultural adaptability of creative content through its unique adaptation method, enhances the participation and satisfaction of different cultural groups, promotes communication and understanding between cultures, and thus achieves effective creative expression and communication in a multicultural environment. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 The present invention is a flowchart of a creative expression dynamic adaptation method based on a multicultural framework.
[0032] Figure 2 It is a schematic diagram of a creative expression dynamic adaptation method based on a multicultural framework according to the present invention.
[0033] Figure 3 It is another schematic diagram of a creative expression dynamic adaptation method based on a multicultural framework according to the present invention. DETAILED DESCRIPTION
[0034] The following will describe the technical solutions in the embodiments of the present invention in detail in conjunction with the accompanying drawings in the embodiments of the present invention. The described embodiments are only part of the embodiments of the present invention.
[0035] In this embodiment, a creative expression dynamic adaptation method based on a multicultural framework may specifically include:
[0036] Step S101, construct a multicultural framework that includes mainstream values, cognitive patterns, and aesthetic preference dimensions, and create a cultural distance matrix to identify the heterogeneity between cultural groups.
[0037] Through social media, user surveys, and online behavior records, we obtain data on users' mainstream values, cognitive patterns, and aesthetic preferences, build a meta-cultural framework that includes multiple cultural characteristics and user behavior dimensions, and determine the importance and interaction methods of each dimension in the framework. Through the pages visited by users, the duration of stay, and the order of clicks, we extract the cultural characteristics of different cultural groups, including language habits, values, and artistic preferences. According to the cultural characteristics of different cultural groups, we use the principal component analysis algorithm to train the model, build a cultural characteristic model, model the cultural characteristics of different cultural groups, and obtain vectors representing the cultural characteristics of different cultural groups. We use the method of calculating the Euclidean distance to calculate the distance between different cultural characteristic vectors, build a cultural distance matrix, and determine the relative position and differences between different cultural groups. According to the cultural distance matrix, we use the K-means clustering algorithm to divide different users into different cultural groups and identify the heterogeneity between cultural groups. We obtain the user's online and offline data, and extract the user's cultural characteristic vector through the established cultural characteristic model. The online data includes browsing history and search habits, and the offline data includes participating cultural activities and geographic location information. According to the distance between the user's cultural characteristic vector and the vectors of different cultural groups, we determine the cultural group to which the user belongs. By continuously tracking user data and monitoring the changes in user cultural feature vectors, a dynamic assessment of user cultural group affiliation can be achieved. For example, data from 100 users were collected through social media, user surveys, and online behavior records. Among them, 50% of users said that the mainstream value is family first, and 50% of users said that the mainstream value is personal freedom. Based on these data, it can be obtained that the number of users who put family first is 50, and the number of users who put personal freedom is 50. Construct a meta-cultural framework that includes multiple cultural features and user behavior dimensions, and determine the importance and interaction mode of each dimension in the framework. Next, the pages visited by users, the dwell time, and the click order are used to extract the cultural features of different cultural groups. After analysis, it is found that the language habits of the family-first group often contain words related to family, children, and parents, while the language habits of the personal freedom group often contain words related to freedom, independence, and individuals. In this way, the cultural feature vectors of the family-first group can be obtained as [6,4], and the cultural feature vectors of the personal freedom group can be obtained as [3,7]. Based on these cultural feature vectors, the distance between different cultural feature vectors can be calculated. By calculating the Euclidean distance, it can be obtained that the distance between the family-first group and the personal freedom group is 5. In this way, a cultural distance matrix can be constructed to determine the relative positions and differences between different cultural groups. Then, based on the cultural distance matrix, the K-means clustering algorithm is used to divide different users into different cultural groups. If the users are divided into two cultural groups, the clustering algorithm can be used to obtain 30 users in the family first group and 70 users in the personal freedom group.In order to further determine the cultural group to which the user belongs, the user's online and offline data are obtained. User A's online data includes browsing history and search habits. The browsing history contains articles about family and children, and the search habits often search for content related to family education. User A's offline data includes cultural activities and geographic location information. He often attends family gatherings and lives in a family-centered community. According to the established cultural feature model, the cultural feature vector of user A can be extracted as [6,4]. By calculating the distance between user A's cultural feature vector and the vectors of different cultural groups, it can be determined that user A belongs to the family-first group. Finally, by continuously tracking user data, the changes in the user's cultural feature vector can be monitored to achieve a dynamic assessment of the user's cultural group affiliation. If user A's online behavior begins to tend towards the characteristics of personal freedom, the cultural feature vector of user A can be updated and the cultural group to which he belongs can be re-evaluated.
[0038] Step S102, designing a dynamic identification mechanism for cultural groups based on the differences in the multicultural framework, and evaluating the cultural distance between different groups in real time.
[0039] Extract the cultural characteristics of each cultural group, including language habits, values, and artistic preferences, and construct a feature vector for each cultural group. If new user data flows in, extract the user's cultural characteristics through the pages visited by the user, the duration of stay, and the order of clicks, and construct a vector representing the user's cultural characteristics. Compare the user's cultural feature vector with the vectors of each cultural group in the framework, and use the Euclidean distance method to calculate the distance between the user vector and the vectors of each cultural group. Set a threshold based on the distance between the user vector and the vectors of each group, and use the threshold to determine the cultural group to which the user is most likely to belong. Continue to obtain new user data and repeat the process of feature extraction, vector calculation, and distance judgment. If the change in the distance between the user vector and the original belonging group vector exceeds the predetermined threshold, it is determined that the user's cultural group has changed, and the user's cultural group affiliation is adjusted. According to the update of all user cultural vectors, evaluate the changes in the members within each cultural group and correct the cultural group feature vector. For example, suppose there are three cultural groups: the American cultural group, the Chinese cultural group, and the Indian cultural group. The cultural characteristics of each cultural group can be extracted in the following ways: Language habits: The American cultural group uses English as the main language, the Chinese cultural group uses Chinese as the main language, and the Indian cultural group uses Hindi as the main language. Values: The American cultural group focuses on personal freedom and the pursuit of happiness, the Chinese cultural group focuses on collective interests and family values, and the Indian cultural group focuses on social hierarchy and religious beliefs. Artistic preferences: The American cultural group prefers pop music and movies, the Chinese cultural group prefers traditional music and drama, and the Indian cultural group prefers Indian music and religious dance. Based on these cultural characteristics, the feature vector of each cultural group can be constructed. For example, the feature vector of the American cultural group can be expressed as [1,0,0], the feature vector of the Chinese cultural group can be expressed as [0,1,0], and the feature vector of the Indian cultural group can be expressed as [0,0,1]. When new user data flows in, the user's cultural characteristics can be extracted through the pages visited by the user, the dwell time, and the click order. Assuming that the user visits the pop music page, stays for a long time, and clicks on a series of popular movies, the user's cultural characteristics can be expressed as [1,0,0], which is similar to the feature vector of the American cultural group. Then, the Euclidean distance method can be used to calculate the distance between the user vector and the vectors of each cultural group. Assume that the Euclidean distance between the user vector and the American cultural group vector is 2, the distance between the user vector and the Chinese cultural group vector is 8, and the distance between the user vector and the Indian cultural group vector is 9. According to the distance between the user vector and each group vector, a threshold value, such as 5, can be set. In this case, the user is most likely to belong to the American cultural group. If new user data is continuously acquired, and the feature extraction, vector calculation, and distance judgment processes are repeated. If the change value of the distance between the user vector and the original belonging group vector exceeds a predetermined threshold (for example, 3), it is determined that the user's cultural group has changed, and the user's cultural group affiliation is adjusted.Based on the update of all user cultural vectors, the changes in the members within each cultural group can be evaluated and the cultural group feature vectors can be corrected. For example, if the cultural feature vectors of a large number of users are close to the feature vectors of the American cultural group, the feature vectors of the American cultural group can be updated to more accurately reflect the cultural characteristics of the group. The above is an example with numerical values, involving calculation, evaluation, analysis, and statistics. In this way, users can be assigned to the most matching cultural group based on their behavior and preferences, and the feature vectors of the group can be continuously adjusted as the data is updated and changed.
[0040] Step S103, constructing a cross-cultural language adaptation model to support dynamic matching and adaptation between creative content expression methods and receiving groups.
[0041] Obtain the cultural feature vector of the target user and determine the cultural group to which he belongs. According to the constructed cultural distance matrix, calculate the cultural distance between the cultural group to which the target user belongs and other cultural groups. According to the cultural distance data, user feature vector and historical feedback data, use the linear regression algorithm to train the model, build a cross-cultural language adaptation model, and predict the degree of adaptation of the target user's cultural group to various creative content expressions. According to the adaptation evaluation results, select the creative content expression with the highest degree of adaptation to the target user's cultural group, match and generate creative content of the corresponding method, and deliver it to the user. After the user browses, clicks, and forwards, obtain the user's feedback on the recommended content. If the user feedback is positive, increase the weight of the corresponding cultural distance and the creative content expression. If the user feedback is negative, reduce the matching weight of the corresponding distance and expression method, and determine the best matching option. By continuously adjusting the weight, optimize the cross-cultural language adaptation model to achieve dynamic matching between the creative content expression method and the receiving group. For example, obtain the cultural feature vector of the target user. For example, the cultural feature vector of the target user is [8,4,6], which indicates the user's preference for the three cultural features. According to the constructed cultural distance matrix, |culture 1, culture 2, culture 3|, |culture 1, 0, 2, 7|, |culture 2, 2, 0, 5|, |culture 3, 7, 5, 0|; calculate the cultural distance between the target user's cultural group and other cultural groups. For example, if the target user's cultural group is culture 1, the cultural distance between it and other cultural groups is calculated to be [0, 2, 7]. According to the cultural distance data, user feature vector and historical feedback data, use the linear regression algorithm to train the model. For example, use the linear regression model to fit the data to obtain the equation, adaptation degree = 2*cultural feature 1+3*cultural feature 2+5*cultural feature 3; according to the adaptation evaluation results, select the creative content expression method with the highest degree of adaptation to the target user's cultural group, for example, the creative content expression method with the highest degree of adaptation is expression method 2. Match and generate creative content of the corresponding method and deliver it to the user. After the user browses, clicks, and forwards, obtain the user's feedback on the recommended content. If the user feedback is positive, increase the weight of the corresponding cultural distance and the creative content expression method, for example, increase the weight of expression method 2 and the cultural distance of 2. If the user feedback is negative, the matching weight of the corresponding distance and expression method is reduced, for example, the weight of expression method 2 and cultural distance 2 is reduced. By continuously adjusting the weights, the cross-cultural language adaptation model is optimized to achieve dynamic matching between the expression method of creative content and the receiving group.
[0042] Step S104, generating a candidate creative content set that takes into account both innovation and diversity based on the generative adversarial network, covering cross-cultural expressions.
[0043] Obtain various existing creative contents as training samples, which cover different creative expressions, including contents from different cultural backgrounds. Obtain labeled training samples, use label information to enhance the samples, increase sample diversity through data expansion technology, and generate a diverse set of candidate creative content. Based on the enhanced data set, build a generative adversarial network and train it, which includes a generative model and a discriminative model. The generative model generates creative content, and the discriminative model evaluates the authenticity and quality of the content. Use the trained generative model and input random noise conditions to generate candidate creative content. If the quality of the generated content is lower than expected, obtain user feedback on some candidate content to evaluate the innovation and diversity of the content. If the user feedback result is not good, further optimize the generative adversarial network based on the feedback data to improve the quality of the content generated by the model. Repeat the generation, evaluation and optimization process until the generated creative content meets the preset innovation and diversity requirements. For example, 1,000 creative content samples from different cultural backgrounds are collected, and each sample is labeled to indicate its innovation and diversity level. The value range of the label is 1 to 10, where 1 indicates low innovation and diversity, and 10 indicates high innovation and diversity. Using these labeled training samples, the sample diversity is increased through data expansion techniques. For each sample, 5 synonym replacement variants are generated, so the dataset will increase by 5,000 samples. Next, the generative adversarial network is used for training. The generative model takes random noise conditions as input and generates candidate creative content. The discriminative model evaluates the authenticity and quality of the content. The adversarial training between the generative model and the discriminative model is used to improve the ability of the generative model. The generative model outputs one creative content each time, and 1,000 training iterations are performed. In each iteration, 100 candidate contents are generated using random noise conditions, and their authenticity and quality are evaluated by the discriminative model. If 20 candidate contents are evaluated as low quality in the 1,000th iteration, these low-quality contents are displayed to users and their feedback is collected. Users give each content a score to indicate its innovation and diversity. The user's score is a range from 0 to 1, where 0 indicates low innovation and diversity and 1 indicates high innovation and diversity. The user's ratings for each low-quality content are collected and the average rating is calculated, resulting in an average rating of 3. Based on user feedback data, we can analyze which aspects lead to the generation of low-quality content and further optimize the generative adversarial network, such as adjusting the network structure of the generative model, changing the distribution of noise input, or adding more training iterations to improve the performance of the generative model. Repeat the process of generation, evaluation, and optimization until the generated creative content meets the preset requirements for innovation and diversity.
[0044] Step S105 , based on the cultural characteristics and interaction data of the users, the distance between cultural groups is evaluated and matching creative content is screened.
[0045] Obtain the user's cultural feature vector, determine the cultural group to which the target user belongs, and extract the feature vector of the cultural group. Use the Euclidean distance method to calculate the cultural distance between the target user vector and the cultural group vector to which it belongs. According to the preset threshold, in the constructed cultural distance matrix, determine the cultural group whose cultural distance with the target user is less than the preset threshold. Based on the generated creative content set, extract the semantic feature vector of each creative content. The semantic features include theme features, emotional features, and style features. Calculate the similarity between each creative content feature vector and the target user group feature vector. If the similarity is higher than the preset threshold, the corresponding creative content is added to the candidate set. According to the feedback data of user browsing, liking, and forwarding, set the reward value of positive feedback. If the user feedback degree is greater than the preset threshold, increase the matching weight of the corresponding creative content and the cultural distance. Through the Q-learning algorithm, adjust the matching weight of the creative content and the cultural distance to optimize the recommendation strategy. Screen out the creative content that matches the cultural distance of the target group and obtains positive feedback from users to form a recommendation set. For example, there is a cultural feature data set containing feature vectors of different cultural groups, including group A, group B, and group C. Each group has a unique feature vector that represents the main characteristics of the culture. Similarly, each user has a feature vector that represents their cultural inclinations. The feature vector of the target user V is [0.8, 0.1, 0.3], and the feature vectors of the cultural groups are, group A: [0.7, 0.2, 0.4], group B: [0.3, 0.8, 0.2], group C: [0.1, 0.1, 0.9]; the Euclidean distance is used to calculate the distance between user V and each group, and the threshold is set to 0.5. The feature vector of user V is closest to group A. Determine similar cultural groups Determine all groups whose cultural distance to user V is less than 0.5. Extract creative content features Assume that we have a creative content library, each content has its own semantic feature vector, and the feature vector of content 1 is obtained as
[0046] [0.6, 0.2, 0.5]. Calculate the similarity between content and user group characteristics. Calculate the similarity between the feature vector of content 1 and the feature vector of group A to which user V belongs. Screen candidate content. If the similarity is higher than the preset threshold of 0.7, add the content to the candidate set. Feedback data and reward value adjust the matching weight of content and cultural distance according to user V's browsing, liking, forwarding and other behaviors. If user V's feedback on content 1 is higher than a certain threshold, increase the weight of the content. Use Q-learning to optimize recommendations. Use the Q-learning algorithm to adjust the matching weight and optimize future recommendation strategies. Form the final recommendation set. Filter out content that matches user V's cultural distance and receives positive feedback to form a recommendation set.
[0047] Based on the analysis of cultural characteristics and user behavior data, a dynamic scoring mechanism for cross-cultural content recommendation is designed and optimized.
[0048] According to cultural characteristics and user behavior patterns, determine the impact of different cultural characteristics on user preferences for recommended content, and construct a comprehensive scoring formula C = α*(F_u-F_c)+β*D+γ*T, where F_u represents the user's cultural feature vector, F_c represents the average feature vector of the cultural group to which the user belongs, D represents the user's behavior data, and T represents the time series data of the user's interaction with the content. α, β, γ are initial weight coefficients set based on theoretical research and adjusted according to actual data. Use API to obtain user cultural feature data, behavior data, and time series data. Perform preprocessing operations such as cleaning and normalization on the obtained data. Use the obtained data to train a recursive neural network model and predict the cultural adaptation score C based on real-time data. Design and implement a real-time user feedback mechanism to obtain user feedback on recommended content. According to the obtained user feedback, adjust the weight coefficients α, β, γ and optimize the scoring formula. According to the optimized scoring formula, dynamically adjust the logic of the recommendation system. Continuously monitor changes in user preferences and the effectiveness of the recommendation system. Implement a continuous learning mechanism for the model to adapt to changes in user behavior and cultural characteristics. Evaluate the model effect regularly and adjust the model based on feedback and analysis results. For example, there is a user N who belongs to cultural group B. It is necessary to predict her cultural adaptation score C for the recommended content based on N's data and make content recommendations accordingly. If N's cultural feature vector F_u is [0.8, 0.5, 0.6], representing the degree of preference for music, movies and food respectively, the average cultural feature vector F_c of group B is [0.7, 0.6, 0.7]. N's behavioral data click-through rate D is 0.75, and the duration of interaction between the time series data and the content T is 30 minutes. Set the initial weight coefficients α = 0.5, β = 0.3, and γ = 0.2. Calculate C = 0.5*((0.8-0.7)+(0.5-0.6)+(0.6-0.7))+0.3*0.75+0.2*30 / 60=0.5*(-0.1)+0.3*0.75+0.2*0.5=-0.05+0.225+0.1=0.275. If N gives positive feedback on a recommendation. Based on the feedback, fine-tune the weight coefficients to α=0.45, β=0.35, γ=0.2. Adjust the algorithm of the recommendation system based on the new weight coefficients and continuously collected user data. Continue to monitor N's response to the new recommendations and continue to optimize the model based on the feedback.
[0049] Step S106, designing a user simulation mechanism to predict the acceptance preferences of different cultural groups and correcting the adaptation error of the creative content expression method.
[0050] Obtain user data from different cultural groups, including browsing and clicking behavior data. Extract vectors representing user characteristics, including points of interest and activity participation; recommend creative content with different expressions to each cultural group, and record user feedback data on the creative content, including browsing and clicking behavior data. Based on the obtained user feedback data, use the decision tree algorithm to establish a mapping model between group user feature vectors and content expression methods, and predict the feedback of users from a certain cultural group on content with a given expression method. In the recommendation process, compare the expression method of the creative content with the user feedback predicted by the model. If there is an adaptation error between the expression method of a certain creative content and the user feedback predicted by the model, select the content expression method with the most positive feedback in the simulation results. Based on the user's own characteristics, use the collaborative filtering algorithm to train the model and generate personalized expression preference predictions. By continuously obtaining user feedback and adjusting the personalized model, the creative content adaptation is continuously optimized, and the user's expression preference is gradually approached to achieve creative content recommendation. For example, the user data of different cultural groups is being studied and vectors representing user characteristics are extracted. The browsing and clicking behavior data of 100 users are collected and represented as binary vectors. User 1: browses web page A, clicks on web page B, and the vector is represented as [1,0,0,1]; User 2: browses web page A, clicks on web page C, and the vector is represented as [1,0,1,0]; User 3: browses web page B, clicks on web page C, and the vector is represented as [0,1,1,0]; Based on the feature vectors of these users, a mapping model between the feature vectors of group users and the content expression methods is established using the decision tree algorithm. If the model predicts the user feature vector [1,0,0,1], the feedback on the content of the given expression method is positive. Now, recommend an expression method of creative content to user 1, and record the user's feedback data, and get that user 1 browsed the content but did not click it. According to the adaptation error between the user feedback predicted by the model and the user's actual feedback, select the content expression method with the most positive feedback in the simulation results. Another expression method in the simulation results predicts that the feedback of the user feature vector [1,0,0,1] is negative. At this time, personalized expression preference prediction is performed, and the collaborative filtering algorithm is used for model training. According to the user's own characteristics, the model predicts that the user likes expression method A. By continuously obtaining user feedback and adjusting the personalized model, we can continuously optimize the adaptation of creative content, gradually approach the user's expression preferences, and achieve accurate creative content recommendations. We can continuously adjust the expression of creative content based on the user's feature vector and model prediction until the user is satisfied.
[0051] Based on cultural characteristics and user behavior analysis, dynamic prediction and adaptation of user preferences in a multicultural environment can be achieved.
[0052] According to the behaviors and preferences of users from different cultural backgrounds, identify the key cultural features that affect users' content preferences, including language habits, aesthetic standards, and values, build a framework that includes multiple cultural features and user behavior dimensions, determine the importance and interaction mode of each dimension in the framework, and design a preliminary prediction formula S = ∑(wi*Ci) + δ(B). Among them, S represents the user's cultural adaptability score for creative content, Ci represents the i-th cultural feature, wi is the weight of the corresponding cultural feature, B represents user behavior data, and δ is the function that processes user behavior data. Obtain relevant data of users from different cultural backgrounds, including the behaviors and feedback of users from different cultural backgrounds. Apply the prediction formula based on the obtained data to test its effectiveness and accuracy. According to the test results, adjust the definition of the cultural feature weight wi and the function δ to optimize the formula. In actual applications, continuously monitor the performance of the formula and obtain real-time feedback from users. According to the new data and feedback obtained, continuously iterate and update the formula to maintain the timeliness and accuracy of the formula. For example, user L comes from a multicultural background and needs to use a prediction formula to evaluate her cultural adaptability to a certain creative content. The key cultural characteristics identified include language habits C1, aesthetic standards C2, and values C3, with initial weights set to w1=0.4, w2=0.3, and w3=0.3. The cultural characteristic data of user L are language habits C1=0.7, aesthetic standards C2=0.8, and values C3=0.6. If the click rate B of user L's behavior data for a specific type of content is 0.85. Set function δ to a simple linear function that processes behavior data, such as δ(B)=2B, and apply function δ to get δ(0.85)=1.7. Apply the prediction formula S=0.4*0.7+0.3*0.8+0.3*0.6+1.7, and get S=0.28+0.24+0.18+1.7=2.4. Verify the accuracy of the formula through user feedback and behavior data analysis. Adjust the weights based on the feedback to obtain new weights of w1 = 0.35, w2 = 0.35, w3 = 0.3, and modify the function δ to obtain a new function of δ(B) = 1.5B. Continue to apply and adjust the formula for newly collected user data and feedback. Recalculate the cultural fit score based on the new data to ensure that the formula remains up-to-date and accurate.
[0053] Step S107, collecting user feedback through A / B testing to evaluate the acceptance effect of creative content in multicultural groups.
[0054] According to the cultural characteristics of users, users are divided into different cultural groups to construct a multicultural user group data set. Samples are randomly selected from the candidate creative content set, and the samples randomly selected from the candidate creative content set are divided into two mutually exclusive test groups, Group A and Group B, according to the proportion of the cultural groups to which the users belong. The traffic of the website or application is redirected to Group A and Group B. The users who visit the content of Group A and Group B represent the corresponding cultural groups respectively. After the user browses the content, the user's behavior data is obtained, including the page dwell time, click-through rate, and conversion rate. According to the user's behavior data, the acceptance effect of Group A / B is calculated. Through the t-test, the acceptance effect of Group A / B is compared to determine whether the difference in the acceptance effect of the two test groups exceeds the preset threshold; if the test results show that the difference in the acceptance effect of different cultural groups on creative content exceeds the preset threshold, it proves that there is a difference in the acceptance effect of different cultural groups on creative content, and the content composition strategy is adjusted according to the acceptance effect of different cultural groups on creative content. Through regular A / B testing, the acceptance effect of each cultural group on creative content is continuously monitored and evaluated, and the creative content generation strategy is continuously revised and optimized according to the test results. For example, for users from different cultural groups, A / B testing is used to evaluate the acceptance of creative content by different cultural groups, and the content generation strategy is optimized based on the test results. Users are divided into two cultural groups, cultural group A and cultural group B, according to their cultural characteristics. A multicultural user group data set is constructed, in which 70% of users belong to cultural group A and 30% of users belong to cultural group B, so that the sample ratio of the test group can be ensured to be consistent with the ratio of the actual user group. Next, a portion of samples are randomly selected from the candidate creative content set as test samples. If 100 samples are selected, these samples are evenly distributed to group A and group B, that is, each group has 50 samples. Then, the traffic of the website or application is redirected to group A and group B, so that users who visit the content of group A and group B represent the corresponding cultural groups respectively. The user's behavior data, including page dwell time, click-through rate and conversion rate, is recorded. In group A, the average page dwell time is 30 seconds, the click-through rate is 10%, and the conversion rate is 5%, while in group B, the average page dwell time is 40 seconds, the click-through rate is 12%, and the conversion rate is 8%. Through the t-test, the acceptance effect of Group A and Group B can be compared, and the preset threshold is set to 5%. The differences between Group A and Group B in terms of page dwell time, click-through rate, and conversion rate are calculated, and then statistical analysis is performed. In terms of page dwell time, the mean of Group A is 30 seconds, the standard deviation is 5 seconds, and the mean of Group B is 40 seconds, and the standard deviation is 6 seconds. Through the t-test, the t value is calculated to be 5, and the significance level is 0.05. Since the t value is greater than the significance level, it can be considered that there is a significant difference between Group A and Group B in terms of page dwell time. Similarly, the differences between Group A and Group B in terms of click-through rate and conversion rate can be compared through the t-test.If the test results show that the difference in the acceptance of creative content by different cultural groups exceeds the preset threshold, it can be concluded that there is a difference in the acceptance of creative content by different cultural groups. The content composition strategy can be adjusted based on the acceptance of creative content by different cultural groups. Therefore, if cultural group A has a better acceptance of a certain creative content, while cultural group B has a poorer acceptance, elements that are more in line with their likes and preferences can be added to the content for cultural group B to improve its acceptance. By conducting A / B tests regularly, the acceptance of creative content by various cultural groups can be continuously monitored and evaluated, and the creative content generation strategy can be continuously revised and optimized based on the test results to meet the needs and preferences of different cultural groups.
[0055] Step S108, adjusting the cultural distance between groups in the multicultural framework based on the feedback effect to achieve dynamic optimization of expression adaptation.
[0056] Obtain the behavior data of users browsing, clicking, and forwarding the recommended creative content. Based on the content style and conversion rate indicators, use the K-means algorithm to train the model and determine the differences in the feedback of users from different cultural groups on the expression of creative content. Based on the full user behavior data, use the method of calculating the Euclidean distance to recalculate the cultural distance between the cultural group vectors. The full user behavior data is the behavior data generated by all users in the process of using the platform or service, including the user's browsing history, click behavior, content preference, forwarding, and social interaction data. After the vector distance is updated, re-evaluate the mapping relationship between users and cultural groups. Based on the new cultural distance, re-evaluate the acceptance of content with different expressions in each cultural group, and form a new matching table between content expression and cultural distance. If the matching weights of certain distances and content expression need to be adjusted, the matching table is updated accordingly. Re-recommend the content and verify the conversion rate of users under the new method. If the conversion rate of users under the new method is improved, the optimization is considered to be effective. Through closed-loop iterative optimization, a dynamic decision table matching cultural distance and expression method is obtained to achieve continuous upgrading of expression adaptation. For example, according to the cultural background of users, the expression method of creative content recommended to them should be optimized. The collected user behavior data include browsing, clicking and forwarding behaviors. The K-means algorithm will be used to cluster these behavior data to determine the differences in feedback from users of different cultural groups on the expression of creative content. First, the behavior data is represented as a vector, where each dimension represents a behavior, including the number of views, clicks, and forwardings. Then, the K-means algorithm is used to divide the users into different clusters, each cluster representing a cultural group. In this way, the center vector of each cluster can be obtained, representing the average feedback of the group on the creative content. Next, the Euclidean distance method is used to recalculate the cultural distance between cultural groups. For example, two cultural groups A and B, their center vectors are vector A(10,5,2) and vector B(8,4,3) respectively. The Euclidean distance between them can be calculated, that is, dist(A,B)=sqrt((10-8) 2 +(5-4) 2 +(2-3) 2 )=sqrt(4+1+1)=sqrt(6)=2.449. Therefore, user U belongs to cultural group A, but is closer to group B in terms of culture distance. Based on the new cultural distance, the expression of the creative content recommended to user U can be adjusted to make it closer to the preferences of group B. Then, the content is recommended to user U again, and it is verified whether the conversion rate is improved. By continuously collecting user behavior data and updating the dynamic decision table that matches cultural distance with expression, the continuous optimization of creative content expression can be achieved.
[0057] Step S109, identifying the taste drift trends of different groups based on user browsing history and feedback data, and optimizing the content recommendation strategy.
[0058] Continuously obtain user feature data, including browsing history, click-through rate, and feedback, to detect changes in user personal tastes. Based on user feature data, use the ARIMA algorithm to train the model, build a taste drift model, and detect fluctuations in user tastes. If it is detected that the user's taste fluctuates, mark the user set with positive and negative taste migration. According to the users whose tastes have changed, explore the distribution patterns of users with different taste migration directions in terms of age and gender. According to the collective drift trend of the tastes of user groups, judge the changing direction of the tastes of different cultural groups in accepting creative content. According to the user's cultural feature vector, screen creative content that matches the user's cultural preferences, and adjust the expression of the selected content. The adjustment of the expression includes adjusting the language style, visual presentation, or theme framework of the content. For users with personalized taste preferences, establish a unique personalized taste model, and in the recommendation process, form a personalized recommendation plan based on the taste of the group to which the user belongs and the user's own personalized taste. After the user browses the creative content, obtain user feedback data to detect the effectiveness of the taste drift model. If the effect deviates, adjust the model parameters according to the feedback data to achieve closed-loop iterative optimization. For example, continuously obtain user feature data, including browsing history, click-through rate, and feedback, to detect changes in user tastes. First, use the ARIMA algorithm to train the model of user feature data and build a taste drift model. Use the ARIMA (1,0,0) model, where 1 represents the order of the autoregressive term. Next, the taste drift model can be used to detect the fluctuation of user tastes. Analyze the feature data of user A, and find that in the recent period, user A's taste has undergone an obvious positive migration, that is, the taste preference has shifted to higher-rated content. User A can be marked as a user with positive taste migration. Further, based on users with changing tastes, the distribution patterns of users with different taste migration directions in terms of age and gender can be explored. It can be found that among users with positive taste migration, users aged between 20 and 30 account for a high proportion, and the gender is mainly male. According to the collective drift trend of the taste of the user group, the direction of change in the taste acceptance of creative content by different cultural groups can be judged. It is found that in a specific cultural group, the taste drift trend evolves in a more open and diversified direction. According to the user's cultural feature vector, creative content that matches the user's cultural preferences is screened, and the expression of the selected content is adjusted. The expression adjustment includes adjusting the language style, visual presentation or theme framework of the content. For users with personalized taste preferences, an exclusive personalized taste model can be established, and in the recommendation process, the taste of the group to which the user belongs and the user's own personalized taste are combined to form a personalized recommendation plan. For example, for user B, it is found that he has a high score for a specific type of creative content, and his taste is quite different from that of most users. Therefore, an exclusive personalized recommendation plan can be provided for user B, customized for his taste.After users browse creative content, user feedback data can be obtained to detect the effectiveness of the taste drift model. User feedback can be compared with the prediction results of the taste drift model. If it is found that there is a deviation between the model's prediction and the user's feedback, the model parameters can be adjusted according to the feedback data to achieve closed-loop iterative optimization and improve the accuracy and effectiveness of the model. In short, by continuously obtaining user feature data and using the ARIMA algorithm to build a taste drift model, it is possible to detect and analyze the fluctuations in user tastes and further explore the taste change patterns of different user groups. By adjusting the content expression method and personalized recommendation scheme, the user's taste needs can be better met, and the closed-loop iterative optimization of the model can be achieved through user feedback data.
[0059] What is disclosed above is only a preferred embodiment of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.
Claims
1. A creative expression dynamic adaptation method based on a multicultural framework, characterized in that: The method comprises: Construct a multicultural framework that includes mainstream values, cognitive patterns, and aesthetic preferences, and create a cultural distance matrix to identify the heterogeneity between cultural groups; design a dynamic identification mechanism for cultural groups based on the differences in the multicultural framework to evaluate the cultural distance between different groups in real time; construct a cross-cultural language adaptation model to support the dynamic matching and adaptation between creative content expression and the receiving group; generate a candidate creative content set that takes into account both innovation and diversity based on the generative adversarial network, covering cross-cultural expressions; evaluate the distance between cultural groups and screen matching creative content based on the user's cultural characteristics and interaction data; design a user simulation mechanism to predict the acceptance preferences of different cultural groups and correct the adaptation errors of creative content expression; collect user feedback through A / B testing to evaluate the acceptance effect of creative content in multicultural groups; adjust the cultural distance between groups in the multicultural framework based on the feedback effect to achieve dynamic optimization of expression adaptation; identify the taste drift trend of different groups based on user browsing history and feedback data, and optimize content recommendation strategies; The construction of a multicultural framework including mainstream values, cognitive patterns, and aesthetic preference dimensions, and the creation of a cultural distance matrix, includes: obtaining user data on mainstream values, cognitive patterns, and aesthetic preferences through social media, user surveys, and online behavior records, constructing a multicultural framework including multiple cultural characteristics and user behavior dimensions, and determining the importance and interaction mode of each dimension in the framework; extracting cultural characteristics of different cultural groups, including language habits, values, and artistic preferences, through the pages visited by users, the dwelling time, and the click sequence; according to the cultural characteristics of different cultural groups, using the principal component analysis algorithm to train the model, construct a cultural characteristic model, model the cultural characteristics of different cultural groups, and obtain vectors representing the cultural characteristics of different cultural groups; using the method of calculating the Euclidean distance to calculate the distance between different cultural characteristic vectors, construct a cultural distance matrix, and determine the relative positions and differences between different cultural groups; The design is based on a dynamic identification mechanism of cultural groups with different multicultural frameworks, and evaluates the cultural distance between different groups in real time, including: Extract the cultural characteristics of each cultural group, including language habits, values and artistic preferences, and construct a feature vector for each cultural group; if new user data flows in, extract the user's cultural characteristics through the pages visited by the user, the dwell time and the click order, and construct a vector representing the user's cultural characteristics; compare the user's cultural feature vector with the vectors of each cultural group in the framework, and use the Euclidean distance method to calculate the distance between the user vector and the vectors of each cultural group; set a threshold based on the distance between the user vector and the vectors of each group, and use the threshold to determine the cultural group to which the user is most likely to belong; continuously obtain new user data, and repeat the feature extraction, vector calculation and distance judgment process; if the change in the distance between the user vector and the original belonging group vector exceeds the predetermined threshold, it is determined that the user's cultural group has changed, and the user's cultural group affiliation is adjusted; based on the update of all user cultural vectors, evaluate the changes in the internal members of each cultural group, and correct the cultural group feature vector; The method of identifying the taste drift trend of different groups based on user browsing history and feedback data includes: obtaining user feature data, the user feature data including browsing history, click-through rate and feedback, and detecting changes in user personal taste; using the ARIMA algorithm to perform model training based on the user feature data, constructing a taste drift model, and detecting fluctuations in user taste; if fluctuations in user taste are detected, marking a set of users with positive and negative taste migration; based on users with changed tastes, mining the distribution patterns of users with different taste migration directions in age and gender dimensions; and judging the direction of changes in the taste acceptance of creative content by different cultural groups based on the collective drift trend of the taste of the user group.
2. The method according to claim 1, wherein: The multicultural framework is constructed to include mainstream values, cognitive patterns, and aesthetic preference dimensions, and a cultural distance matrix is created to identify heterogeneity between cultural groups, including: Based on the cultural distance matrix, the K-means clustering algorithm is used to divide different users into different cultural groups and identify the heterogeneity between cultural groups. The user's online and offline data are obtained, and the user's cultural feature vector is extracted through the established cultural feature model. The online data includes browsing history and search habits, and the offline data includes cultural activities participated in and geographic location information. The cultural group to which the user belongs is determined based on the distance between the user's cultural feature vector and the vectors of different cultural groups. By continuously tracking user data and monitoring the changes in the user's cultural feature vector, a dynamic evaluation of the user's cultural group affiliation is achieved.
3. The method according to claim 1, wherein: The cross-cultural language adaptation model is constructed to support dynamic matching and adaptation between the creative content expression method and the receiving group, including: Obtain the cultural feature vector of the target user and determine the cultural group to which he belongs; calculate the cultural distance between the target user's cultural group and other cultural groups based on the constructed cultural distance matrix; use the linear regression algorithm to train the model based on the cultural distance data, user feature vector and historical feedback data, build a cross-cultural language adaptation model, and predict the degree of adaptation of the target user's cultural group to various creative content expressions; according to the adaptation evaluation results, select the creative content expression method with the highest degree of adaptation to the target user's cultural group, match and generate creative content of the corresponding method, and deliver it to the user; obtain user feedback on the recommended content after the user browses, clicks and forwards; if the user feedback is positive, increase the weight of the corresponding cultural distance and creative content expression method; If the user feedback is negative, the matching weight of the corresponding distance and expression method will be lowered to determine the best matching option; by continuously adjusting the weight, the cross-cultural language adaptation model will be optimized to achieve dynamic matching between the expression method of creative content and the receiving group.
4. The method according to claim 1, wherein: The candidate creative content set based on the generative adversarial network is generated with both innovation and diversity, covering cross-cultural expressions, including: Obtain various existing creative content as training samples, covering different creative expressions, including content from different cultural backgrounds; Obtain labeled training samples, use label information to enhance the samples, increase sample diversity through data expansion technology, and generate a diverse set of candidate creative content; Based on the enhanced data set, build and train a generative adversarial network, which includes a generative model and a discriminative model. The generative model generates creative content, and the discriminative model evaluates the authenticity and quality of the content; Use the trained generative model and input random noise conditions to generate candidate creative content; If the quality of the generated content is lower than expected, obtain user feedback on some of the candidate content to evaluate the innovation and diversity of the content; If the user feedback results are not good, the generative adversarial network will be further optimized based on the feedback data to improve the quality of the content generated by the model; the process of generation, evaluation and optimization will be repeated until the generated creative content meets the preset innovation and diversity requirements.
5. The method according to claim 1, wherein: The method of evaluating the distance between cultural groups and screening matching creative content based on the cultural characteristics and interaction data of users includes: Obtain the user's cultural feature vector, determine the cultural group to which the target user belongs, and extract the feature vector of the cultural group; use the Euclidean distance method to calculate the cultural distance between the target user vector and the cultural group vector to which it belongs; according to the preset threshold, determine the cultural group whose cultural distance with the target user is less than the preset threshold in the constructed cultural distance matrix; based on the generated creative content set, extract the semantic feature vector of each creative content, the semantic features include theme features, emotional features and style features; Calculate the similarity between the feature vector of each creative content and the feature vector of the target user group; if the similarity is higher than the preset threshold, add the corresponding creative content to the candidate set; set the reward value for positive feedback based on the feedback data of user browsing, liking and forwarding; if the user feedback level is higher than the preset threshold, increase the matching weight of the corresponding creative content and cultural distance; Through the Q-learning algorithm, the matching weight between creative content and cultural distance is adjusted to optimize the recommendation strategy; Filter out creative content that matches the cultural distance of the target group and has received positive feedback from users to form a recommendation set; It also includes: designing and optimizing a dynamic scoring mechanism for cross-cultural content recommendations based on the analysis of cultural characteristics and user behavior data.
6. The method according to claim 1, wherein: The user simulation mechanism is designed to predict the acceptance preferences of different cultural groups and correct the adaptation errors of creative content expression methods, including: Obtain user data from different cultural groups, including browsing and clicking behavior data; Extract vectors representing user characteristics, including points of interest and activity participation; recommend creative content with different expressions to different cultural groups, and record user feedback data on creative content, including browsing and clicking behavior data; use decision tree algorithm to establish a mapping model between group user feature vectors and content expression methods based on the obtained user feedback data, and predict the feedback of users of a certain cultural group on content with a given expression method; in the recommendation process, compare the expression method of creative content with the user feedback predicted by the model. If there is an adaptation error between the expression method of a certain creative content and the user feedback predicted by the model, select the content expression method with the most positive feedback in the simulation results; use collaborative filtering algorithm to train the model based on the user's own characteristics to generate personalized expression preference predictions; By continuously obtaining user feedback and adjusting the personalized model, we continuously optimize the adaptation of creative content, gradually approach the user's expression preferences, and realize creative content recommendation; it also includes: based on cultural characteristics and user behavior analysis, we realize dynamic prediction and adaptation of user preferences in a multicultural environment.
7. The method according to claim 1, wherein: The A / B test method is used to collect user feedback and evaluate the acceptance of creative content in multicultural groups, including: According to the cultural characteristics of users, users are divided into different cultural groups to construct a multicultural user group data set; samples are randomly selected from the candidate creative content set, and the samples randomly selected from the candidate creative content set are divided into two mutually exclusive test groups, Group A and Group B, according to the proportion of the cultural groups to which the users belong; the traffic of the website or application is redirected to Group A and Group B, and the users who visit the content of Group A and Group B represent the corresponding cultural groups respectively; after the users browse the content, the user's behavior data is obtained, including the page dwell time, click-through rate and conversion rate; based on the user's behavior data, the acceptance effect of Group A / B is calculated; through the t-test, the acceptance effect of Group A / B is compared to determine whether the difference in the acceptance effect of the two test groups exceeds the preset threshold; if the test results show that the difference in the acceptance effect of different cultural groups on creative content exceeds the preset threshold, it proves that there is a difference in the acceptance effect of different cultural groups on creative content, and the content composition strategy is adjusted according to the acceptance effect of different cultural groups on creative content; through regular A / B testing, the acceptance effect of each cultural group on creative content is continuously monitored and evaluated, and the creative content generation strategy is continuously revised and optimized according to the test results.
8. The method according to claim 1, wherein: The method of adjusting the cultural distance between groups in a multicultural framework based on feedback effects to achieve dynamic optimization of expression adaptation includes: Obtain user behavior data on browsing, clicking, and forwarding recommended creative content; use the K-means algorithm to train the model based on content style and conversion rate indicators to determine the differences in feedback from users of different cultural groups on the expression of creative content; recalculate the cultural distance between cultural group vectors based on the full user behavior data using the Euclidean distance calculation method. The full user behavior data refers to the behavior data generated by all users in the process of using the platform or service, including users' browsing history, click behavior, content preference, forwarding, and social interaction data; After the vector distance is updated, the mapping relationship between users and cultural groups is re-evaluated; based on the new cultural distance, the acceptance of content with different expressions in various cultural groups is re-evaluated to form a new matching table between content expression and cultural distance; if the matching weights of certain distances and content expressions need to be adjusted, the matching table is updated accordingly; the content is recommended again and the user conversion rate under the new method is verified. If the user conversion rate under the new method is improved, the optimization is considered to be effective; through closed-loop iterative optimization, a dynamic decision table matching cultural distance and expression method is obtained to achieve continuous upgrading of expression adaptation.
9. The method according to claim 1, wherein: The method of identifying the taste drift trends of different groups based on user browsing history and feedback data and optimizing content recommendation strategies includes: According to the user's cultural feature vector, creative content that matches the user's cultural preferences is screened, and the expression of the selected content is adjusted. The adjustment of the expression includes adjusting the language style and visual presentation or theme framework of the content; for users with personalized taste preferences, a unique personalized taste model is established, and in the recommendation process, a personalized recommendation plan is formed based on the taste of the user's group and the user's own personalized taste; After users browse creative content, obtain user feedback data to test the effectiveness of the taste drift model; If the effect deviates, the model parameters are adjusted according to the feedback data to achieve closed-loop iterative optimization.
Citation Information
Patent Citations
User dynamic classification-based e-commerce platform commodity recommendation method and system
CN111709812A
Push resource determination method based on big data mining and cloud computing AI service system
CN113468404A
