Real-time assisted decision-making and wording suggestion system for cross-cultural communication scenario
By combining multimodal input fusion and contextual reasoning techniques with deep neural networks and two-layer cultural semantic graphs, personalized cross-cultural communication suggestions are generated, solving the problem of lack of real-time scene awareness and personalized guidance in existing systems, and achieving more efficient cross-cultural communication support.
Patent Information
- Application Number
- PCT/CN2025/142183
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-11-12
AI Technical Summary
Existing cross-cultural communication systems lack real-time scenario awareness and personalized guidance, making them unable to adapt to dynamic communicative situations, resulting in mechanical and poorly adaptable suggestions.
It employs a multimodal input fusion module, a cultural semantic graph and contextual reasoning module, a communication strategy and language suggestion generation module, and an interactive feedback module, combined with deep neural networks and a two-layer cultural semantic graph, to perceive and generate personalized language suggestions in real time.
It improves the efficiency and accuracy of cross-cultural communication, provides highly flexible and intelligent decision support, avoids the limitations of fixed models, and adapts to real-time support in different cultural contexts.
Smart Images

Figure CN2025142183_12112026_PF_FP_ABST
Abstract
Description
A real-time decision support and language suggestion system for cross-cultural communication scenarios Technical Field
[0001] This invention belongs to the field of cross-cultural human-computer interaction technology, and in particular relates to a real-time auxiliary decision-making and language suggestion system for cross-cultural communication scenarios. Background Technology
[0002] As globalization deepens, cross-cultural exchanges are becoming increasingly frequent. In cross-border business, diplomacy, education, and social settings, cultural differences often lead to communication barriers and even conflicts. Existing technologies largely rely on static cultural knowledge bases or offline training materials, failing to adapt to real-time, dynamic communication environments. Traditional cross-cultural training tools lack context awareness and immediate intervention capabilities, making it difficult for users to obtain targeted guidance in specific situations.
[0003] In recent years, some studies have attempted to provide cross-cultural language suggestions using natural language processing (NLP) technology. However, most systems rely solely on simple keyword matching or fixed dialogue templates, lacking a comprehensive understanding of cultural context, nonverbal behavior, and communicative dynamics. Existing solutions often overlook the real-time nature of scenarios, individual user differences, and the dynamic evolution of cultural rules, resulting in mechanical and poorly adaptable suggestions that cannot cope with the complex and ever-changing needs of real-world communication. Furthermore, existing technologies tend to focus on the linguistic level, failing to integrate multimodal information and real-time contextual analysis, thus limiting the system's practicality and response accuracy.
[0004] To address these issues, we provide a real-time decision support and language suggestion system for cross-cultural communication scenarios. Technical issues
[0005] To address the shortcomings of existing technologies, this invention provides a real-time decision support and language suggestion system for cross-cultural communication scenarios. The technical problem this invention aims to solve is: how to overcome the static, mechanical, and unsuitable-for-changing communication scenarios of existing technologies by combining cultural semantic graphs with contextual reasoning and personalized feedback mechanisms. Technical solutions
[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution.
[0007] This invention is a real-time decision support and language suggestion system for cross-cultural communication scenarios, including a multimodal input fusion module. The multimodal input fusion module fuses input signals from the communication scenario to form multidimensional perceptual information. The input signals include speech signals, facial images, posture data, and semantic context.
[0008] The cultural semantic graph and contextual reasoning module performs contextual reasoning processing on the multidimensional perceptual information to generate reasoning results. The cultural semantic graph and contextual reasoning module includes a two-layer cultural semantic graph and a context dependency analyzer.
[0009] A communication strategy and language suggestion generation module generates comprehensive communication strategies and basic language suggestions based on the reasoning results.
[0010] A multi-level output module generates multi-dimensional output based on the comprehensive communication strategy and the basic language suggestions. The multi-level output module includes a semantic output layer, a tone output layer, and a non-verbal prompt layer.
[0011] An interactive feedback module receives user feedback and generates personalized word suggestions based on multi-dimensional output personalized analysis of the user feedback.
[0012] The present invention is further configured such that the fusion processing employs a deep neural network, the deep neural network comprising a feature extraction layer, a temporal modeling layer, and a modal fusion layer, wherein the feature extraction layer extracts the spatial features of the input signal, the temporal modeling layer models the temporal relationship of the input signal, and the modal fusion layer performs weighted fusion of the spatial features to generate the multidimensional sensing information.
[0013] The present invention is further configured such that the two-layer cultural semantic graph includes a static general cultural knowledge layer and a dynamic conversational context layer. The static general cultural knowledge layer is stored in a graph structure and includes cultural concept nodes, social norm nodes, taboo custom nodes, and typical behavior pattern nodes. The dynamic conversational context layer is constructed in real time according to the current dialogue and includes the identity of the current participant, the entity of the topic being discussed, and relevant cultural concepts obtained from the static general cultural knowledge layer.
[0014] The present invention is further configured such that the steps of the contextual reasoning process are:
[0015] S1. The context dependency analyzer parses the multidimensional perceptual information to generate a parsing result, which includes semantic entities, sentiment tags, and behavioral features. The parsing result and the cultural concept node are then matched for similarity to generate a cultural concept activation set.
[0016] S2. Starting from the nodes in the cultural concept activation set, perform path traversal in the static general cultural knowledge layer to generate a set of context-related nodes. The path traversal involves extracting the social norm nodes, taboo custom nodes, and typical behavior pattern nodes corresponding to the cultural concept activation set in the static general cultural knowledge layer. The set of context-related nodes, the current participant's identity identifier, and the entity of the topic being discussed are used as common nodes to update the dynamic conversation context layer to obtain a new dynamic conversation context layer.
[0017] S3. The new dynamic conversational context layer combines the multidimensional perceptual information to generate the reasoning result.
[0018] The present invention is further configured such that the reasoning result includes a cultural fit assessment report and situational adaptation feedback.
[0019] The present invention is further configured such that the communication strategy and terminology suggestion generation module includes a strategy parsing unit, a rule base query unit, and a personalized adaptation engine. The strategy parsing unit receives the reasoning result, the rule base query unit stores a cross-cultural communication strategy template library and a terminology pattern library, the rule base query unit retrieves a communication strategy framework matching the reasoning result from the strategy template library, and retrieves a terminology template that adapts to the reasoning result from the terminology pattern library, and the personalized adaptation engine performs weighted adjustment on the communication strategy framework and the terminology template to generate the comprehensive communication strategy and the basic terminology suggestion.
[0020] The present invention is further configured such that the multi-dimensional output includes semantic dimension output, tone dimension output, and behavior dimension output. The semantic output layer performs multi-step transformation processing on the basic terminology suggestions to generate the semantic dimension output. The tone output layer performs parameterized adjustment on the user's actual speech features based on the cultural adaptability assessment in the inference result and the style tendency in the comprehensive communication strategy to generate the tone dimension output. The non-verbal cue layer performs normative adaptation on the user's non-verbal behavior features to generate the behavior dimension output.
[0021] The present invention is further configured such that the multi-step conversion process includes calling the static general cultural knowledge layer in the cultural semantic graph and contextual reasoning module to retrieve synonyms or near-synonyms that conform to the current cultural context and replacing them.
[0022] The present invention is further configured such that the personalized analysis step is as follows:
[0023] S1. Generate adjustment rules based on the differences between the actual user expressions reflected in the user feedback and the multi-dimensional output.
[0024] S2. Reinterpret the reasoning results according to the adjustment rules to generate the individualized terminology suggestions.
[0025] The present invention is further configured such that the individualized language suggestions include personalized semantic expression schemes, adaptive tone control guidelines, and customized non-verbal behavior prompts. Beneficial effects
[0026] The beneficial effects of this invention are as follows: By combining multimodal input fusion and contextual reasoning technologies, this invention can perceive and analyze various signals in communicative scenarios in real time, and generate dynamic cultural adaptation strategies based on these signals. Unlike traditional template-based communication systems, this invention provides personalized and adaptive language suggestions by deeply understanding the cultural background, emotional state, and context of communicative participants, thereby improving the efficiency and accuracy of cross-cultural communication.
[0027] This invention constructs a two-layer cultural semantic graph, dynamically updates the association between cultural concepts and social norms, and generates strategy suggestions most suitable for the current communication scenario based on real-time dialogue. This highly flexible and intelligent decision support system avoids the limitations of fixed patterns and static rules, greatly improving the system's adaptability and practicality, and providing accurate and effective real-time support for communication in different cultural contexts. Attached Figure Description
[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below.
[0029] Figure 1 is a diagram of the overall system architecture of the present invention.
[0030] Figure 2 is a schematic diagram of the two-layer cultural semantic graph and contextual reasoning mechanism of the present invention.
[0031] Figure 3 is a flowchart of the personalized analysis and suggestion generation process of the present invention. Embodiments of the present invention
[0032] The technical solutions of the present invention will be described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments.
[0033] Example 1
[0034] Please refer to Figures 1-3. This invention is a real-time auxiliary decision-making and language suggestion system for cross-cultural communication scenarios, including a multimodal input fusion module. This module fuses input signals from the communication scenario to form multidimensional perceptual information. The input signals include speech signals, facial images, posture data, and semantic context. The fusion processing employs a deep neural network, which includes a feature extraction layer, a temporal modeling layer, and a modality fusion layer. The feature extraction layer extracts spatial features from the input signals, the temporal modeling layer models the temporal relationships of the input signals, and the modality fusion layer performs weighted fusion of the spatial features to generate multidimensional perceptual information.
[0035] Feature extraction:
[0036] The input signals include speech signals, facial images, pose data, and semantic context. Feature extraction is performed on the signals in the following ways:
[0037] Speech signal: Time-spectral features are extracted using a convolutional neural network. The time-spectral graph of the input speech signal is a 256×256 matrix. The convolutional neural network extracts spatial feature vectors representing the speech through multiple convolution operations. A 1×512 feature vector is obtained after extraction.
[0038] Facial image: The facial image is processed using a deep convolutional neural network. The input image size is 224×224×3. After extracting facial features through the deep convolutional neural network, a 1×512 feature vector is obtained.
[0039] Pose data: Sensor data is processed using a 3D convolutional neural network. The pose data for each frame is a 50×50×3 matrix. Through 3D convolution operations, a 1×256 feature vector is obtained.
[0040] Semantic context: A pre-trained BERT model is used to extract text features. The input text is "How are you today?", and a 1×768 feature vector is obtained through the BERT model.
[0041] The feature vectors represent the input signals of each modality, and a preliminary high-dimensional representation is generated through the feature extraction layer.
[0042] Temporal modeling:
[0043] The temporal dependencies of the input signal are processed using a Long Short-Term Memory (LSTM) network. The feature vector of each mode is processed through an LSM layer:
[0044] Speech features are processed through a Long Short-Term Memory (LSTM) network, and the model outputs a 1×512 vector containing temporal information. Facial and pose features are processed separately through LSM networks, each generating a 1×512 vector. Semantic context is processed using BERT's temporal modeling capabilities, resulting in a 1×512 vector. These temporally modeled feature vectors are then passed to the modality fusion layer for weighted fusion.
[0045] Modal fusion:
[0046] The task of the modality fusion layer is to perform weighted fusion of data from different modalities.
[0047] The weighting coefficients w1, w2, w3, and w4 are automatically learned during training through backpropagation and gradient descent. The initial weighting coefficients are: w1=0.4, w2=0.3, w3=0.2, and w4=0.1.
[0048] After weighted fusion, the fusion characteristics are obtained:
[0049] Fusion features = 0.4 × speech features + 0.3 × facial features + 0.2 × posture features + 0.1 × semantic features.
[0050] The result of temporal modeling for the feature vectors of each modality is as follows:
[0051] Voice features: [0.5, 0.3, 0.4], facial features: [0.6, 0.2, 0.1], posture features: [0.1, 0.5, 0.3], semantic features: [0.7, 0.2, 0.4].
[0052] The weighted fusion calculation is as follows:
[0053] Fusion feature = 0.4 × [0.5, 0.3, 0.4] + 0.3 × [0.6, 0.2, 0.1] + 0.2 × [0.1, 0.5, 0.3] + 0.1 × [0.7, 0.2, 0.4].
[0054] The result is:
[0055] Fusion features = [0.42, 0.33, 0.36].
[0056] The fused features represent weighted multimodal information, which is used for subsequent decision-making or word suggestion generation.
[0057] The Cultural Semantic Graph and Contextual Reasoning module processes multi-dimensional perceptual information to generate reasoning results. This module includes a two-layer cultural semantic graph and a context-dependent analyzer. The two-layer cultural semantic graph consists of a static general cultural knowledge layer and a dynamic conversational context layer. The static general cultural knowledge layer is stored in a graph structure and includes cultural concept nodes, social norm nodes, taboo custom nodes, and typical behavior pattern nodes. The dynamic conversational context layer is constructed in real-time based on the current dialogue and includes the identities of the current participants, the entities involved in the discussion, and relevant cultural concepts obtained from the static general cultural knowledge layer.
[0058] The steps of situational reasoning are as follows:
[0059] S1. The context dependency analyzer parses multidimensional perceptual information to generate parsing results, which include semantic entities, sentiment labels, and behavioral features. The parsing results are then matched with cultural concept nodes to generate a cultural concept activation set.
[0060] S2. Starting with the nodes in the cultural concept activation set, perform path traversal in the static general cultural knowledge layer to generate a set of context-related nodes. The path traversal involves extracting social norm nodes, taboo custom nodes, and typical behavior pattern nodes corresponding to the cultural concept activation set from the static general cultural knowledge layer. The set of context-related nodes, the current participant's identity, and the entity of the topic being discussed are used as common nodes to update the dynamic conversation context layer, resulting in a new dynamic conversation context layer.
[0061] S3. The new dynamic conversational context layer combines multi-dimensional perceptual information to generate inference results. The inference results include a cultural fit assessment report and contextual adaptation feedback.
[0062] The communication strategy and language suggestion generation module generates comprehensive communication strategies and basic language suggestions based on the reasoning results. This module includes a strategy parsing unit, a rule base query unit, and a personalized adaptation engine. The strategy parsing unit receives the reasoning results. The rule base query unit stores a cross-cultural communication strategy template library and a language pattern library. The rule base query unit retrieves communication strategy frameworks matching the reasoning results from the strategy template library and language templates that fit the reasoning results from the language pattern library. The personalized adaptation engine performs weighted adjustments on the communication strategy frameworks and language templates to generate comprehensive communication strategies and basic language suggestions.
[0063] The multi-level output module generates multi-dimensional output based on comprehensive communication strategies and basic language suggestions. This module includes a semantic output layer, a tone output layer, and a non-verbal cue layer. The multi-dimensional output comprises semantic, tone, and behavioral dimensions. The semantic output layer performs multi-step transformation processing on the basic language suggestions to generate semantic dimension output. The tone output layer, based on the cultural fit assessment in the inference results and the style tendency in the comprehensive communication strategies, parametrically adjusts the user's actual speech characteristics to generate tone dimension output. The non-verbal cue layer performs normative adaptation on the user's non-verbal behavioral characteristics to generate behavioral dimension output. The multi-step transformation processing includes calling the cultural semantic graph and the static general cultural knowledge layer in the contextual reasoning module to retrieve synonyms or near-synonyms that conform to the current cultural context and replacing them.
[0064] The interactive feedback module receives user feedback and generates personalized wording suggestions based on multi-dimensional analysis of this feedback. The personalized analysis steps are as follows:
[0065] S1. Generate adjustment rules based on the differences between the actual user expression reflected in user feedback and the multi-dimensional output.
[0066] S2. Reinterpret the inference results according to the adjustment rules to generate personalized language suggestions. Personalized language suggestions include personalized semantic expression schemes, adaptive tone control guidelines, and customized non-verbal behavior prompts.
[0067] Example 2
[0068] Please refer to Figures 1-2. Based on Example 1, a two-layer cultural semantic graph and a context-dependent analyzer are used to perform context-based reasoning on multi-dimensional perceptual information to generate real-time communication strategies and language suggestions based on cultural adaptability and contextual feedback.
[0069] 1. Two-layer cultural semantic graph structure
[0070] The two-layer cultural semantic graph consists of a static general cultural knowledge layer and a dynamic conversational context layer, with the following functions and structure:
[0071] Static general cultural knowledge layer: Stored in a graph structure, containing fixed cultural concepts and social rules that do not change with dialogue. It includes the following nodes:
[0072] Cultural concept nodes: such as etiquette, rituals, and face-saving culture. Social norm nodes: such as social behavioral norms like respecting elders and hosting meals. Taboo custom nodes: such as sensitive topics that cannot be mentioned in certain situations. Typical behavioral pattern nodes: such as social behavior in formal occasions and social behavior in informal occasions.
[0073] Dynamic conversation context layer: Built based on real-time changes in the current dialogue, containing information related to the dialogue context:
[0074] Current participant identity identifier: Indicates the identity of the person involved in the current conversation, such as customer, waiter, leader, etc. Current topic of discussion entity: The current topic of discussion is meeting arrangements or holiday travel.
[0075] Relevant cultural concepts obtained from the static general cultural knowledge layer: such as cultural habits or behavioral norms related to the current topic.
[0076] 2. Situational Reasoning Processing Steps
[0077] 2.1 Context Dependency Analyzer Parses Multidimensional Perceptual Information
[0078] The context-dependent analyzer parses the multidimensional perceptual information and generates the parsing results. The multidimensional perceptual information includes input signals such as speech, facial expressions, and posture.
[0079] The analysis results include:
[0080] Semantic entities: Specific content extracted from input speech or text, such as meeting times or travel arrangements. Sentiment labels: Analyzing the emotional tone of the dialogue, such as positive or negative. Behavioral features: Such as body language or tone of voice, reflecting the speaker's behavioral attitude.
[0081] The parsed results are matched with cultural concept nodes based on similarity. By calculating the similarity value between each parsed result and a cultural concept node, a cultural concept activation set is generated. The cultural concept activation set contains the most relevant cultural concepts in the current dialogue.
[0082] 2.2 Path traversal generates a set of context-related nodes.
[0083] Each node in the cultural concept activation set is used as a starting point for path traversal. Path traversal extracts other nodes related to the activated node from the static general cultural knowledge layer, forming a set of context-related nodes. For example:
[0084] If the cultural concept node is meeting arrangements, then the social norm nodes involved include punctuality or meeting etiquette. It may also involve taboo customs, such as avoiding discussing personal matters. Typical behavioral patterns might include communication methods in formal settings.
[0085] The set of context-related nodes, the current participant's identity, and the entity of the topic being discussed are used as common nodes to update the dynamic conversation context layer, forming a new dynamic conversation context layer. The update operation adjusts the context layer according to the participants and discussion topic of the current dialogue, ensuring that the new dynamic conversation context layer reflects the real-time changes in the context.
[0086] 2.3 Generating reasoning results by combining multidimensional sensory information
[0087] The new dynamic conversational context layer combines multi-dimensional perceptual information to generate inference results.
[0088] The inference results include:
[0089] Cultural Adaptability Assessment Report: This report assesses the degree of compatibility between the communication strategies employed in the current communication scenario and the target cultural background. Based on information from cultural concept nodes and contextual layers, it evaluates whether the communication behavior conforms to specific cultural requirements.
[0090] Context-adaptive feedback: Provides adaptive suggestions for the current conversation to ensure smooth communication. Offers appropriate tone, intonation, or nonverbal cues based on the participants' identities and emotional labels.
[0091] 3. Data Example
[0092] The input signals for the current dialogue include the following:
[0093] Voice signal: "Is the meeting you'd like to schedule about this year's budget discussion?"
[0094] Facial expression: Smiling, indicating friendliness. Posture data: Leaning forward, indicating active participation.
[0095] The analysis results are as follows:
[0096] Semantic entities: meeting, budget, discussion; sentiment label: positive; behavioral characteristics: leaning forward, smiling. After similarity matching, the generated cultural concept activation set includes meeting scheduling and punctuality.
[0097] During the path traversal step, the static general cultural knowledge layer extracts social norm nodes and typical behavior pattern nodes related to the meeting arrangements, and updates the new dynamic conversation context layer by combining the current participant identities and discussion topic entities.
[0098] Combining multidimensional perceptual information and a new dynamic conversational context layer, the generated inference results include:
[0099] Cultural Adaptability Assessment Report: Current communicative behaviors are well-matched with the cultural background of the meeting and are highly adaptable.
[0100] Context-adaptive feedback: It is recommended to maintain a formal tone when speaking and to ensure that budget questions are answered accurately.
[0101] Example 3
[0102] Please refer to Figure 3. Based on Examples 1 and 2, by analyzing user feedback, communication strategies and language suggestions are dynamically adjusted to generate personalized communication output that meets user needs and optimize the interactive experience.
[0103] 1. Receive user feedback
[0104] User feedback primarily stems from the interaction process, where users offer opinions on the tone, wording, or manner of expression in suggestions. In one example, user feedback is as follows:
[0105] User feedback: The tone is too formal; it needs to be more natural. User feedback: Specific suggestions are needed.
[0106] 2. Generate adjustment rules based on user feedback
[0107] Based on user feedback, the interaction feedback module will perform personalized analysis and generate adjustment rules. The specific analysis process is as follows:
[0108] Reflecting differences and generating adjustment rules:
[0109] Difference identification: By comparing the user's actual feedback with the system's multi-dimensional output, the system identifies any discrepancies. The system found that the user wanted a more natural tone and therefore needed to adjust it.
[0110] Generate adjustment rules: Based on feedback, generate adjustment rules. If the user requests a natural tone, the formal tone should be adjusted to a friendly tone, and interjections should be added. For vague feedback, generate rules to specify the vague suggestions and provide clear action steps.
[0111] 3. Reinterpret the reasoning results and generate personalized terminology suggestions.
[0112] Adjusting the inference results: Based on the generated adjustment rules, the interactive feedback module will reinterpret the inference results and generate personalized terminology suggestions based on the adjusted rules. The specific steps are as follows:
[0113] Tone Adjustment: Replace the formal tone in the reasoning result with a friendly tone. If the reasoning result is "Please be sure to attend the meeting on time," then change it to "Can we discuss the meeting schedule together?"
[0114] Specific suggestions: For vague user feedback, the system will provide more specific suggestions based on adjustment rules. The original inference result was "Let's discuss the budget issue," which has been adjusted to "In which areas would you like to adjust the budget?"
[0115] Behavioral suggestion adjustments: If user feedback indicates that certain nonverbal behaviors are unsuitable for the current context, the system will optimize behavioral suggestions based on adjustment rules. If feedback mentions that excessive body language is unnecessary, the system will suggest reducing unnecessary gestures and replacing them with more natural eye contact.
[0116] 4. Suggestions for individualized terminology output
[0117] Based on the steps above, personalized terminology suggestions include:
[0118] Personalized semantic expression scheme: Adjusted semantic expression methods. Adaptive tone control guidelines: Adjust tone based on user feedback to make the text tone better meet user expectations. Customized non-verbal behavior prompts: Personalized adjustments to non-verbal behavior based on user feedback.
[0119] Based on the above user feedback, the suggested personalized wording is as follows:
[0120] Semantic expression scheme: "Can we discuss the meeting schedule together?"
[0121] Tone control guidelines: Use a friendly and open tone, avoiding excessive formality. Nonverbal behavior tips: Maintain appropriate eye contact and avoid excessive gestures.
[0122] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.
Claims
1. A real-time decision support and language suggestion system for cross-cultural communication scenarios. Includes, characterized in that: A multimodal input fusion module fuses input signals from a communicative scenario to form multidimensional perceptual information. The input signals include speech signals, facial images, posture data, and semantic context. The cultural semantic graph and contextual reasoning module performs contextual reasoning processing on the multidimensional perceptual information to generate reasoning results. The cultural semantic graph and contextual reasoning module includes a two-layer cultural semantic graph and a context dependency analyzer. A communication strategy and language suggestion generation module generates comprehensive communication strategies and basic language suggestions based on the reasoning results; A multi-level output module generates multi-dimensional output based on the comprehensive communication strategy and the basic language suggestions. The multi-level output module includes a semantic output layer, a tone output layer, and a non-verbal prompt layer. An interactive feedback module receives user feedback and generates personalized word suggestions based on multi-dimensional output personalized analysis of the user feedback.
2. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The fusion process employs a deep neural network, which includes a feature extraction layer, a temporal modeling layer, and a modal fusion layer. The feature extraction layer extracts the spatial features of the input signal, the temporal modeling layer models the temporal relationship of the input signal, and the modal fusion layer performs weighted fusion of the spatial features to generate the multidimensional sensing information.
3. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The dual-layer cultural semantic graph includes a static general cultural knowledge layer and a dynamic conversational context layer. The static general cultural knowledge layer is stored in a graph structure and includes cultural concept nodes, social norm nodes, taboo custom nodes, and typical behavior pattern nodes. The dynamic conversational context layer is constructed in real time based on the current dialogue and includes the identity of the current participant, the entity of the topic being discussed, and relevant cultural concepts obtained from the static general cultural knowledge layer.
4. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 3, characterized in that: The steps of the contextual reasoning process are as follows: S1. The context dependency analyzer parses the multidimensional perceptual information to generate a parsing result, which includes semantic entities, sentiment tags and behavioral features. The parsing result is then matched with the cultural concept nodes to generate a cultural concept activation set. S2. Starting from the nodes in the cultural concept activation set, perform path traversal in the static general cultural knowledge layer to generate a set of context-related nodes. The path traversal involves extracting the social norm nodes, taboo custom nodes, and typical behavior pattern nodes corresponding to the cultural concept activation set in the static general cultural knowledge layer. The set of context-related nodes, the current participant's identity identifier, and the entity of the topic being discussed are used as common nodes to update the dynamic conversation context layer to obtain a new dynamic conversation context layer. S3. The new dynamic conversational context layer combines the multidimensional perceptual information to generate the reasoning result.
5. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The reasoning results include a cultural fit assessment report and situational adaptation feedback.
6. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The communication strategy and terminology suggestion generation module includes a strategy parsing unit, a rule base query unit, and a personalized adaptation engine. The strategy parsing unit receives the reasoning result. The rule base query unit stores a cross-cultural communication strategy template library and a terminology pattern library. The rule base query unit retrieves a communication strategy framework that matches the reasoning result from the strategy template library and retrieves a terminology template that matches the reasoning result from the terminology pattern library. The personalized adaptation engine performs weighted adjustments on the communication strategy framework and the terminology template to generate the comprehensive communication strategy and the basic terminology suggestions.
7. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The multi-dimensional output includes semantic dimension output, tone dimension output, and behavioral dimension output. The semantic output layer performs multi-step transformation processing on the basic terminology suggestions to generate the semantic dimension output. The tone output layer parameterizes the user's actual speech features based on the cultural fit assessment in the inference results and the style tendency in the comprehensive communication strategy to generate the tone dimension output. The non-verbal cue layer performs normative adaptation on the user's non-verbal behavioral features to generate the behavioral dimension output.
8. The real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 7, characterized in that: The multi-step conversion process includes calling the static general cultural knowledge layer in the cultural semantic graph and the contextual reasoning module to retrieve synonyms or near-synonyms that conform to the current cultural context and replacing them.
9. A real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The steps of the personalized analysis are as follows: S1. Generate adjustment rules based on the differences between the actual user expressions reflected in the user feedback and the multi-dimensional output; S2. Reinterpret the reasoning results according to the adjustment rules to generate the individualized terminology suggestions.
10. A real-time decision support and language suggestion system for cross-cultural communication scenarios according to claim 1, characterized in that: The individualized language suggestions include personalized semantic expression schemes, adaptive tone control guidelines, and customized non-verbal behavior prompts.
