Big data-based hotpot seasoning flavor intelligent adaptation method and system

Abstract

The application discloses a hot pot seasoning flavor intelligent adaptation method and system based on big data, relates to the technical field of food flavor intelligent recommendation, and discloses the hot pot seasoning flavor intelligent adaptation method and system based on big data, which constructs a heterogeneous graph by integrating multi-modal user data and environmental factors, generates accurate flavor recommendation in combination with a dynamic prediction model, effectively solves the defects that a traditional method cannot analyze user segmentation demand and ignores environmental influence, and can improve the adaptation accuracy of hot pot seasoning flavor and users.

Description

Technical Field

[0001] This application relates to the field of intelligent recommendation technology for food flavors, and in particular to a method and system for intelligent adaptation of hot pot base flavors based on big data. Background Technology

[0002] With the increasing popularity of hot pot dining and the growing trend of personalized consumer demand, the precise adaptation of hot pot broth flavor has become a key technological direction for improving user experience. Existing methods mainly rely on user-selected preset tags (such as "mildly spicy" or "medium numbing") or rating data, combined with algorithms such as collaborative filtering and rule-based matching for recommendations. However, the above solutions have the following technical shortcomings: First, massive amounts of unstructured user feedback data (such as detailed flavor descriptions, taste experiences, and subjective emotional expressions in reviews) cannot be effectively mined and quantified. Due to the strong subjectivity and complex and variable semantics of these data, traditional technologies struggle to accurately map them to a computable flavor dimension space, resulting in systems that can only capture rough preferences and cannot analyze users' dynamic needs for specific features such as the duration of numbing sensation, the impact of spiciness, and the layers of umami. Second, existing technologies lack the ability to integrate multimodal data, failing to effectively integrate flavor feature information from multiple data sources such as text reviews, voice feedback, and historical behavioral data. Furthermore, existing recommendation systems generally ignore the impact of environmental factors (such as season and temperature) on flavor perception, resulting in recommendations that lack contextual adaptability and fail to meet users' needs for refined and dynamic flavor adaptation under different environmental conditions. In addition, when users' flavor requirements exceed the flavor range of pre-stored base ingredients, existing systems lack the ability to automatically generate customized recipe suggestions, leading to damage to the integrity and continuity of the user experience.

[0003] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main purpose of this application is to provide a method and system for intelligent adaptation of hot pot base flavor based on big data, aiming to improve the accuracy of the adaptation of hot pot base flavor to users.

[0005] To achieve the above objectives, this application proposes a method for intelligent flavor adaptation of hot pot base based on big data, the method comprising:

[0006] Acquire multimodal user data; the multimodal user data includes text data of user comments, preset taste preference tag data, user historical behavior data, speech-to-text data, and environmental data including season and temperature;

[0007] Natural language processing is performed on the text data and speech-to-text data of the user comments to obtain the identified flavor description words and related intensity words and sentiment words;

[0008] Based on the identified flavor descriptive words and related intensity and sentiment words, combined with the preset flavor preference tag data, an analytical model that integrates flavor domain knowledge is used to process the data to obtain the user flavor perception vector; the user flavor perception vector contains quantified values ​​of multiple predefined flavor feature dimensions;

[0009] Construct a heterogeneous graph containing user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes. Based on the user's historical behavior data, the user's flavor perception vector, the environmental data, and a pre-stored basic flavor perception vector library containing basic flavor perception vectors corresponding to each hot pot base, establish the association relationships between user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes in the heterogeneous graph.

[0010] Based on the heterogeneous spectrum, the historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data, the target flavor perception vector expected by the user is predicted.

[0011] Calculate the similarity distance between the basic flavor perception vector of each hot pot base in the basic flavor perception vector library and the target flavor perception vector;

[0012] Based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base, a hot pot base recommendation list is generated and output.

[0013] In one embodiment, the step of obtaining a user's flavor perception vector by processing the identified flavor descriptive words and related intensity and sentiment words, combined with the preset flavor preference tag data, using an analytical model that integrates flavor domain knowledge, includes:

[0014] By identifying and filtering flavor descriptive words and related intensity and sentiment words that effectively involve predefined flavor feature dimensions through a pre-built ontology library of hot pot flavor domain, effective flavor description data and related effective intensity and sentiment data can be obtained.

[0015] The effective flavor description data is processed using a pre-trained language model to generate a flavor description semantic vector containing word semantic information.

[0016] The effective intensity data and effective sentiment data are mapped to intensity quantization values ​​and sentiment quantization values ​​of corresponding predefined flavor feature dimensions;

[0017] Based on the flavor description semantic vector, intensity quantification value, and sentiment quantification value, combined with preset flavor preference tag data, the user flavor perception vector is generated.

[0018] In one embodiment, the step of converting the valid flavor description data into its flavor description semantic vector in the flavor semantic space includes:

[0019] The effective flavor description data is processed using a pre-trained language model as the basic encoder, and higher weights are assigned to words related to predefined flavor feature dimensions during the self-attention calculation process of the pre-trained language model to enhance the representation ability of flavor semantics, thereby outputting the flavor description semantic vector.

[0020] In one embodiment, the step of mapping the effective intensity data and effective sentiment data to intensity quantization values ​​and sentiment quantization values ​​corresponding to predefined flavor feature dimensions includes:

[0021] A lightweight classifier is used to identify the intensity level of effective intensity data in a predefined flavor feature dimension, and the identified intensity level is converted into a corresponding numerical quantification value, which is then used as the intensity quantification value.

[0022] A lightweight classifier is used to identify the emotional polarity category and its intensity level of effective emotional data in a predefined flavor feature dimension, and the identified emotional polarity category and intensity level are converted into corresponding numerical quantification values ​​as the emotional quantification values.

[0023] In one embodiment, the step of establishing the association between user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes in the heterogeneous graph includes:

[0024] Based on the user's historical behavior data, establish a preference relationship edge between the user node and the hot pot base node;

[0025] Based on the quantized values ​​of each flavor feature dimension in the user's flavor perception vector, a perception intensity relationship edge is established between the user node and the flavor feature dimension node, and the perception intensity relationship edge has a weight representing the quantized value.

[0026] Based on the specific quantized values ​​of each flavor feature dimension in the basic flavor perception vector of each hot pot base in the basic flavor perception vector library, a strong relationship edge is established between the hot pot base node and the flavor feature dimension node. The strong relationship edge has a weight representing the quantized value of the hot pot base in each flavor feature dimension.

[0027] Based on environmental data including seasons and temperature, and combined with the changes in user flavor preferences under different environments obtained by analyzing the correlation between user historical behavior data and historical environmental data, an influence relationship edge is established between environmental data nodes and user nodes on the perceived intensity of specific flavor feature dimensions. The influence relationship edge is used to represent the potential influence pattern of the environment on the user's expected changes in flavor perception.

[0028] In one embodiment, the step of predicting the user's desired target flavor perception vector based on the heterogeneous spectrum, a historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data includes:

[0029] The heterogeneous graph is processed using a graph neural network model. The input includes user nodes, historical user flavor perception vector nodes associated with the user nodes through perception strength relationship edges, hot pot base nodes associated with the user nodes through preference relationship edges, environmental data nodes that the user nodes have historically established associations with, and information about the user's current environment. By propagating and aggregating information along the relationship edges between nodes on the heterogeneous graph, the dynamic influence pattern of environmental changes on user flavor expectations is learned, and the target flavor perception vector is finally output.

[0030] In one embodiment, the step of calculating the similarity distance between the basic flavor perception vector of each hot pot base ingredient in the basic flavor perception vector library and the target flavor perception vector includes:

[0031] Based on the user preference pattern reflected by the user flavor perception vector or the historical user flavor perception vector sequence, or the importance weight information actively set by the user, the importance weight of each flavor feature dimension is determined.

[0032] The weighted distance metric is used to calculate the weighted difference value of the predicted target flavor perception vector and the basic flavor perception vector of each hot pot base in the basic flavor perception vector library in each flavor feature dimension as the similarity distance.

[0033] In one embodiment, the step of generating and outputting a hot pot base recommendation list based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base includes:

[0034] The similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base is used to determine the similarity distance. If the similarity distance between the basic flavor perception vector of all hot pot bases in the basic flavor perception vector library and the predicted target flavor perception vector exceeds a preset mismatch threshold, then based on the target flavor perception vector, a pre-trained regression model is used to deduce the target physicochemical index combination requirements. The regression model is used to characterize the mapping relationship between the flavor perception vector and the key physicochemical indicators of the hot pot base, such as capsaicin content, Sichuan pepper content, and free amino acid composition. This regression model is trained based on historically accumulated sensory evaluation data and physicochemical test data of hot pot bases.

[0035] Based on the derived target physicochemical index combination requirements, combined with the pre-stored ingredient library and pre-stored ratio rule library, a custom base sauce recipe suggestion that meets the target physicochemical index combination requirements is generated through an optimization algorithm, and the custom base sauce recipe suggestion is included in the final output hot pot base sauce recommendation list.

[0036] In one embodiment, the method further includes:

[0037] Receive user feedback data on the hot pot base recommendations in the list; the feedback data includes newly generated comment text data, speech-to-text data, rating data, or purchase behavior data generated by the user for the recommended base they tried;

[0038] The feedback data is used as part of new multimodal user data. Combined with the environmental data when the feedback data was acquired, the process returns to the step of performing natural language processing on the text data and speech-to-text data of the user comments to obtain the identified flavor description words and related intensity words and sentiment words.

[0039] Furthermore, to achieve the above objectives, this application also proposes a big data-based intelligent adaptation system for hot pot base flavors. The big data-based intelligent adaptation system for hot pot base flavors includes: a memory, a processor, and a big data-based intelligent adaptation program for hot pot base flavors stored in the memory and executable on the processor. The big data-based intelligent adaptation program for hot pot base flavors is configured to implement the steps of the big data-based intelligent adaptation method for hot pot base flavors.

[0040] The big data-based intelligent adaptation method and system for hot pot base flavor proposed in this application integrates multimodal user data and environmental factors to construct a heterogeneous spectrum, and combines it with a dynamic prediction model to generate accurate flavor recommendations. This effectively solves the shortcomings of traditional methods that cannot analyze user segmentation needs and ignore environmental influences, and can improve the accuracy of hot pot base flavor adaptation to users. Attached Figure Description

[0041] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating an embodiment of the intelligent flavor adaptation method for hot pot base materials based on big data in this application.

[0044] Figure 2 For this application Figure 1 Detailed flowchart of step S300;

[0045] Figure 3 For this application Figure 1 Detailed flowchart of step S400;

[0046] Figure 4 This is a schematic diagram of a structure provided for an embodiment of the intelligent flavor adaptation system for hot pot base based on big data in this application.

[0047] Explanation of icon numbers:

[0048] 10. Memory; 20. Processor.

[0049] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0050] The technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application, but merely represents selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0051] It should be understood that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0052] In existing technologies, hot pot base recommendations primarily rely on user-selected preset tags or rating data, combined with collaborative filtering or rule matching algorithms. However, existing methods cannot effectively handle the complex semantic information in unstructured user feedback data, such as detailed flavor descriptions, taste experiences, and emotional expressions in reviews. Due to the lack of quantitative analysis of subdivided flavor characteristics, the system can only capture rough preferences and struggles to identify dynamic user needs regarding the duration of numbing sensation and the intensity of spiciness. Furthermore, the influence of environmental factors on flavor perception is not incorporated into the recommendation model, resulting in a lack of scenario adaptability. For example, in cold winter conditions, users may prefer higher-spiciness bases, but traditional systems cannot automatically adjust their recommendation strategies.

[0053] To address these issues, the inventors identified three core shortcomings in existing technologies: low utilization of unstructured data, insufficient precision in quantifying flavor dimensions, and lack of modeling of environmental factors. Analysis of user reviews revealed that most feedback contained combinations of intensity and sentiment terms; for example, complex descriptions like "spicy for a long time but not intoxicating" were difficult to parse using traditional tagging systems. Further research showed that user ratings for the same base sauce fluctuated significantly across different seasons, indicating that environmental variables significantly influence flavor preferences. Based on this, the inventors proposed constructing a multimodal data processing framework that integrates semantic parsing, domain knowledge graphs, and dynamic environmental modeling to achieve precise flavor adaptation.

[0054] Based on this, the embodiments of this application provide a method for intelligent adaptation of hot pot base flavor based on big data, referring to... Figure 1 The big data-based intelligent flavor adaptation method for hot pot base includes steps S100 to S700, wherein:

[0055] Step S100: Obtain multimodal user data; the multimodal user data includes text data of user comments, preset taste preference tag data, user historical behavior data, speech-to-text data, and environmental data including season and temperature;

[0056] Step S200: Perform natural language processing on the text data and speech-to-text data of the user comments to obtain the identified flavor description words and related intensity words and sentiment words;

[0057] Step S300: Based on the identified flavor descriptive words and related intensity and sentiment words, and combined with the preset flavor preference tag data, the user flavor perception vector is obtained by using an analytical model that integrates flavor domain knowledge; the user flavor perception vector contains quantized values ​​of multiple predefined flavor feature dimensions.

[0058] Step S400: Construct a heterogeneous graph containing user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes. Based on the user's historical behavior data, the user's flavor perception vector, the environmental data, and a pre-stored basic flavor perception vector library containing basic flavor perception vectors corresponding to each hot pot base, establish the association relationship between user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes in the heterogeneous graph.

[0059] Step S500: Based on the heterogeneous spectrum, the historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data, predict the target flavor perception vector expected by the user.

[0060] Step S600: Calculate the similarity distance between the basic flavor perception vector of each hot pot base in the basic flavor perception vector library and the target flavor perception vector.

[0061] Step S700: Based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base, generate and output a hot pot base recommendation list.

[0062] In this embodiment, multimodal user data refers to a composite dataset formed by integrating user reviews, preset tags, historical behaviors, speech-to-text conversion, and environmental parameters. Specifically, it can be achieved by collecting e-commerce platform review data through API interfaces and obtaining ambient temperature data from mobile device sensors. This data integration provides multidimensional information support for subsequent analysis. Natural language processing refers to word segmentation, entity recognition, and sentiment analysis of text and speech data. Specifically, it can use the BERT model combined with a domain dictionary to enhance flavor entity recognition capabilities. This processing can extract key descriptions such as "numbing sensation" and "spiciness" and their modifiers. User flavor perception vector refers to the quantitative representation of flavor descriptions mapped to predefined dimensions through an analytical model. Specifically, it can be achieved by fusing semantic vectors and intensity values ​​using a multilayer perceptron. This vector can accurately represent the intensity of a user's perception of specific flavor features. Heterogeneous graph refers to establishing a multidimensional relationship network between users, base ingredients, flavor dimensions, and environmental nodes. Specifically, it can use a graph database to store node attributes and relation edge weights. This structure can effectively capture the dynamic correlation between environmental factors and flavor preferences.

[0063] In this embodiment, the system first collects data generated by users from multiple channels, such as text reviews on e-commerce platforms, voice feedback within the app, historical purchase records, and seasonal information obtained from device location. An NLP model is used to parse flavor descriptions such as "fresh and rich" and "lingering spiciness," and intensity modifiers such as "very" and "slightly." Using a pre-built hot pot flavor ontology library, "lingering spiciness" is mapped to a numbing and spiciness dimension. Combined with positive reviews obtained from sentiment analysis, a user perception vector containing quantitative values ​​for each dimension is generated. Next, a heterogeneous graph is constructed, connecting user nodes with preferred base ingredient nodes, and establishing strong correlation edges between user nodes and flavor dimension nodes such as "numbing" and "spiciness." After introducing environmental nodes, historical data analysis reveals an increased demand for numbing in winter, establishing a positive correlation edge between ambient temperature and the numbing dimension. When predicting the target flavor vector, the graph neural network propagates information along the edges from the temperature node to the numbing dimension, dynamically adjusting recommendation weights. Finally, the cosine similarity between the flavor vectors of each product in the base ingredient library and the target vector is calculated, outputting a recommendation list sorted by matching degree.

[0064] In this embodiment, the proposed solution overcomes the limitations of traditional recommendation systems that rely solely on structured labels. It achieves deep analysis of unstructured feedback through multimodal data processing. Furthermore, the heterogeneous spectrum constructed in this solution incorporates environmental variables into the flavor perception model for the first time, enabling the capture of the dynamic impact of temperature changes on user needs. Simultaneously, this solution achieves finer-grained matching through flavor vector similarity calculation, overcoming the technical deficiency of traditional methods in distinguishing between "persistent numbing" and "instant spiciness." Through the above technical solutions, this application achieves three core effects: First, it transforms users' subjective descriptions into quantifiable flavor dimensions, solving the problem of low utilization of unstructured data; second, it models the dynamic relationship between environmental factors and flavor preferences through heterogeneous spectrum modeling, making the recommendation results adaptive to different scenarios; third, it utilizes similarity calculation to achieve precise matching of base flavor characteristics, improving the fine-grained decision-making capability of the recommendation system. In practical applications, the system can accurately recommend base recipes containing high capsaicin but low volatile oils based on complex descriptions in user reviews such as "wanting spiciness without irritating the throat," and automatically increases the spiciness level of the recommended base in cold weather.

[0065] In one feasible implementation, refer to Figure 2 Step S300 includes steps S310 to S340, wherein:

[0066] Step S310: Identify and filter the content that effectively involves the predefined flavor feature dimensions in the flavor description words and related intensity words and sentiment words through the pre-built hot pot flavor domain ontology library, so as to obtain effective flavor description data and related effective intensity data and effective sentiment data;

[0067] Step S320: Process the effective flavor description data using a pre-trained language model to generate a flavor description semantic vector containing word semantic information;

[0068] Step S330: Map the effective intensity data and effective sentiment data to the intensity quantization value and sentiment quantization value of the corresponding predefined flavor feature dimension;

[0069] Step S340: Based on the flavor description semantic vector, intensity quantization value, and sentiment quantization value, and combined with preset flavor preference tag data, generate the user flavor perception vector.

[0070] In this embodiment, the hot pot flavor domain ontology refers to a pre-established structured knowledge base containing hot pot flavor-related concepts and their relationships. Specifically, it can be implemented using flavor feature dimensions and associated vocabulary sets defined by expert experience, used to filter effective descriptive words related to the predefined flavor dimensions from the raw data. The pre-trained language model refers to a natural language processing model trained on large-scale text data. Specifically, it can be implemented using a BERT model based on the Transformer architecture, enhancing the ability to capture flavor semantics by adjusting the weights of flavor-related words in the self-attention mechanism. Intensity quantification refers to the process of converting intensity words into numerical indicators. Specifically, it can be implemented using numerical mapping rules based on the output of an intensity level classifier, for example, mapping "slight," "moderate," and "strong" to 0.3, 0.6, and 0.9 respectively. Sentiment quantification refers to the process of converting sentiment words into numerical indicators. Specifically, it can be implemented using numerical mapping rules based on the output of a sentiment polarity classifier, for example, mapping "dislike," "neutral," and "like" to -1, 0, and +1 respectively.

[0071] In this embodiment, the flavor descriptions extracted from user reviews and voice data may contain a large number of unstructured expressions, such as "the spiciness lasts a long time" or "the numbing sensation is not obvious enough." After filtering out effective words related to predefined dimensions such as "numbing duration" and "spiciness intensity" using a hot pot flavor ontology, a pre-trained language model is used to generate semantic vectors for descriptions such as "duration" and "not obvious enough." At the same time, "long time" is mapped to an intensity value of 0.8, and "not enough" is mapped to a sentiment value of -0.5. Combined with the user's preset "prefer high numbing" tag, the semantic vectors and quantified values ​​are weighted and fused to finally generate a user flavor perception vector containing dimensions such as "numbing duration = 0.75, spiciness intensity = 0.6".

[0072] In this embodiment, the proposed solution accurately filters flavor-related descriptions using a domain ontology library and separates intensity and sentiment information using a language model with adjusted weights. This achieves fine-grained analysis of unstructured flavor descriptions, enabling the application to effectively extract the subtle flavor requirements implicit in user reviews and transform subjective descriptions into computable quantifiable vectors. By distinguishing between intensity and sentiment as two types of quantifiable indicators, the solution avoids the misjudgment caused by traditional methods that simply categorize "insufficient numbness" as negative sentiment, thus improving the accuracy of user flavor preference representation and providing precise input data for subsequent recommendation algorithms.

[0073] In one feasible implementation, the step of converting the effective flavor description data into a flavor description semantic vector in the flavor semantic space includes: processing the effective flavor description data using a pre-trained language model as a basic encoder, and assigning higher weights to words related to predefined flavor feature dimensions during the self-attention calculation process of the pre-trained language model to enhance the representation ability of flavor semantics, thereby outputting the flavor description semantic vector.

[0074] In this embodiment, the pre-trained language model refers to a neural network model pre-trained on large-scale text data. Specifically, it can be implemented using models with contextual understanding capabilities, such as BERT or RoBERTa, to extract semantic information from flavor description data. The self-attention computation process refers to the process within the model that generates semantic representations by calculating the weights of associations between words. This can be implemented using a multi-head attention mechanism, adjusting the weight distribution to make the model pay more attention to words related to flavor features. Words related to predefined flavor feature dimensions refer to words that have semantic associations with pre-defined flavor dimensions such as numbing sensation, spiciness, and umami. For example, "Sichuan pepper" corresponds to numbing sensation, and "chili pepper" corresponds to spiciness. This can be identified by constructing a flavor domain dictionary or using clustering algorithms.

[0075] In this embodiment, when encoding effective flavor description data, the model's self-attention layer incorporates prior knowledge of predefined flavor feature dimensions, assigning higher attention weights to words related to specific flavor dimensions. For example, when the input text contains "rich aroma of Sichuan peppercorns," the model, when calculating the attention weight of the word "Sichuan peppercorns," enhances the semantic contribution of that word in the numbing sensation dimension by using externally injected information related to the numbing sensation feature. This weight adjustment allows the model to focus more on key descriptive words related to flavor features when generating semantic vectors, thereby enhancing the discriminative power of the vectors in the flavor dimension space.

[0076] In this embodiment, the proposed solution adjusts the self-attention weight distribution to enable the model to actively enhance the semantic expression of flavor-related words during the encoding process. This avoids the problem of flavor feature information being diluted by irrelevant words in the general semantic vector, achieving accurate conversion from flavor description text to flavor semantic vectors. This solves the problem of inaccurate user flavor preference parsing caused by semantic representation bias in existing technologies. By enhancing the semantic weights of flavor-related words, the generated vectors can more completely retain the user's perceptual features in subdivided dimensions such as spiciness and numbness, providing highly discriminative input data for subsequent recommendation algorithms.

[0077] In one feasible implementation, the step of mapping the effective intensity data and effective sentiment data to intensity quantification values ​​and sentiment quantification values ​​of corresponding predefined flavor feature dimensions includes: identifying the intensity level of the effective intensity data on the predefined flavor feature dimension using a lightweight classifier, and converting the identified intensity level into the corresponding numerical quantification value as the intensity quantification value; identifying the sentiment polarity category and its intensity level of the effective sentiment data on the predefined flavor feature dimension using a lightweight classifier, and converting the identified sentiment polarity category and intensity level into the corresponding numerical quantification value as the sentiment quantification value.

[0078] In this embodiment, the lightweight classifier refers to a machine learning model with low computational resource consumption and suitable for real-time processing. Specifically, it can be implemented using a decision tree-based or logistic regression model. Its function is to quickly classify and identify intensity words and sentiment words. The intensity level refers to dividing intensity words into multiple discrete levels, for example, assigning values ​​1, 2, and 3 to "mild," "medium," and "extra spicy," respectively. Its function is to convert the intensity of subjective descriptions into calculable numerical values. The sentiment polarity category refers to dividing sentiment words into positive, neutral, and negative categories, with the intensity level further refined into weak, medium, and strong levels. For example, mapping "very much like" to a positive-strong level and corresponding to a value of +3 , its function is to quantify the user's emotional inclination towards specific flavor characteristics.

[0079] In this embodiment, during implementation, words from the effective intensity data, such as "slightly numbing" and "very spicy," are input into a pre-trained lightweight classifier. This classifier outputs the corresponding intensity level according to predefined intensity level classification rules. For example, when "very spicy" is input, the classifier identifies it as the third intensity level and converts it to the value 3. Simultaneously, words from the effective sentiment data, such as "dislike," are input into another lightweight classifier. This classifier first determines its sentiment polarity to be negative, then identifies its intensity level as moderate, and finally converts it to the value -2. These quantified values ​​are integrated into the user's flavor perception vector, providing structured input data for subsequent recommendation algorithms.

[0080] Compared to existing technologies, traditional methods typically use fixed rules or manual annotation to process intensity and sentiment words, such as directly mapping "spicy" to a fixed numerical value, making it impossible to distinguish between "mildly spicy" and "extra spicy." This solution, however, automatically identifies multi-level intensity and sentiment levels through a machine learning model, more accurately reflecting subtle differences in users' perception of flavor. It achieves refined quantification of unstructured user feedback data, solving the problem in existing technologies where intensity and sentiment descriptions cannot be effectively mapped to a numerical space. This allows the recommendation system to accurately capture users' dynamic needs for flavor characteristics such as numbing sensation and spiciness, thereby enhancing the personalization of hot pot base recommendations.

[0081] In one feasible implementation, refer to Figure 3 Step S400 includes steps S410 to S440, wherein:

[0082] Step S410: Based on the user's historical behavior data, establish a preference relationship edge between the user node and the hot pot base node;

[0083] Step S420: Based on the quantized values ​​of each flavor feature dimension in the user flavor perception vector, establish a perception intensity relationship edge between the user node and the flavor feature dimension node, wherein the perception intensity relationship edge has a weight representing the quantized value.

[0084] Step S430: Based on the specific quantization values ​​of each flavor feature dimension in the basic flavor perception vector of each hot pot base in the basic flavor perception vector library, establish a strong relationship edge between the hot pot base node and the flavor feature dimension node. The strong relationship edge has a weight representing the quantization value of the hot pot base in each flavor feature dimension.

[0085] Step S440: Based on environmental data including season and temperature, and combined with the user's flavor preference change pattern under different environments obtained by analyzing the correlation between user historical behavior data and historical environmental data, an influence relationship edge is established between environmental data nodes and user nodes on the perceived intensity of specific flavor feature dimensions. The influence relationship edge is used to represent the potential influence pattern of the environment on the user's expected changes in flavor perception.

[0086] In this embodiment, the preference relationship edge refers to the association between a user and hot pot base constructed using historical user behavior data. Specifically, edge weights can be established using user purchase records or rating data; for example, the more times a user purchases, the higher the edge weight. The perceived intensity relationship edge refers to the quantified perceived intensity of a user's specific flavor characteristics. Specifically, the weight value can be obtained by normalizing the values ​​of the corresponding dimensions in the user's flavor perception vector. The strength relationship edge refers to the quantified intensity of the hot pot base in a specific flavor characteristic dimension. Specifically, it can be obtained by standardizing the values ​​of the corresponding dimensions in the basic flavor perception vector. The influence relationship edge refers to the potential impact pattern of environmental factors on changes in user flavor preferences. Specifically, it can be modeled using statistical patterns of user preference changes under the same environmental conditions in historical data.

[0087] In this embodiment, the preference relationship edge between the user node and the hot pot base node is constructed by parsing the user's historical order data. For example, if a user has purchased a certain beef tallow hot pot base three times, an edge with a weight of 3 is established between the user node and the corresponding base node. The perception intensity relationship edge between the user node and the flavor feature dimension node is generated by the user's flavor perception vector. For example, if a user scores 0.8 in the "numbing" dimension, an edge with a weight of 0.8 is established between the user node and the "numbing" node. The strength relationship edge between the hot pot base node and the flavor feature dimension node is generated by pre-stored basic flavor perception vectors. For example, if a certain clear broth base scores 0.9 in the "umami" dimension, an edge with a weight of 0.9 is established between the corresponding nodes. The association edge between the environmental data node and the user node is established by analyzing the correlation between environmental parameters and changes in user preferences in historical data. For example, if a user's preference for "spiciness" increases by 20% when the temperature is below 10℃, an influence relationship edge between the temperature node and the user node in the "spiciness" dimension is established.

[0088] In this embodiment, the heterogeneous graph constructed by this solution can simultaneously capture the complex interactions of user preferences, product characteristics, flavor dimensions, and environmental influences through the association relationships of four types of nodes. For example, the increased preference for spicy flavors by users in low-temperature winter environments can be quantitatively expressed through the association edges between environmental nodes and user nodes. Thus, this application effectively solves the problem that existing technologies cannot dynamically reflect the impact of environmental factors on user preferences, achieving deep fusion of multi-dimensional data. Specifically, when low-temperature environmental data is input, the system can automatically increase the recommendation weight of spicy base ingredients, while in high-temperature environments, it increases the ranking position of refreshing base ingredients, making the recommendation results environmentally adaptive. Simultaneously, by quantifying the weight settings of the relationship edges, it can accurately reflect the differences in the intensity of user preferences for different flavor characteristics. For example, the difference between a user's preference for "numbing intensity" (0.8) and "freshness" (0.5) can be accurately expressed through edge weights, avoiding recommendation bias caused by the coarse-grained matching of traditional labeling systems.

[0089] In one feasible implementation, the step of predicting the target flavor perception vector expected by the user based on the heterogeneous graph, the historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data includes: processing the heterogeneous graph using a graph neural network model, taking into input information including user nodes, historical user flavor perception vector nodes associated with the user nodes through perception intensity relationship edges, hot pot base nodes associated with the user nodes through preference relationship edges, environmental data nodes that the user nodes have historically established associations with, and the user's current environmental data; learning the dynamic influence pattern of environmental changes on user flavor expectations by propagating and aggregating information along the relationship edges between nodes on the heterogeneous graph; and finally outputting the target flavor perception vector.

[0090] In this embodiment, referring to the above description, a heterogeneous graph refers to a multi-type node network structure including user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes. Specifically, it can be implemented using a graph database or graph structure data model to express the relationships between user preferences, base attributes, flavor dimensions, and environmental factors. A graph neural network model refers to a deep learning model capable of processing heterogeneous graphs. Specifically, it can be implemented using a heterogeneous graph attention network or a relational graph convolutional network, capturing the dynamic influence of environmental factors on user flavor expectations through the information transmission mechanism of relational edges between nodes. Information propagation and aggregation refers to the process of transmitting and fusing feature information between nodes in a graph neural network. Specifically, it can be implemented using an attention mechanism or message passing algorithm, modeling the relationship patterns between the environment and user preferences by dynamically adjusting the weights of different relational edges.

[0091] In this embodiment, user nodes in the heterogeneous graph are associated with historical flavor perception vectors through perception intensity relationship edges and with hot pot base nodes through preference relationship edges. Simultaneously, influence relationship edges are established between environmental data nodes and user nodes. The graph neural network compares current environmental data with historical environmental data, transmitting environmental change information to user nodes along influence relationship edges. It also aggregates features of users' historical hot pot base selections along preference relationship edges and integrates the trend of flavor perception changes under different environments along perception intensity relationship edges. Through iterative updates of node features across multiple network layers, a target flavor perception vector reflecting the combined influence of the current environment and historical behavior is finally generated at the user node.

[0092] In this embodiment, the solution constructs a heterogeneous graph to explicitly model the multidimensional relationships between users, base ingredients, flavor dimensions, and environment. It utilizes a graph neural network to capture the potential influence paths between environmental changes and user behavior, enabling the prediction results to dynamically adapt to changes in external conditions such as seasons and temperatures. This allows the application to automatically adjust the prediction model of user flavor expectations based on environmental changes, solving the problem of poor scenario adaptability of recommendation results caused by ignoring environmental factors in existing technologies. This achieves more accurate hot pot base ingredient recommendations for users under different seasons or temperature conditions.

[0093] In one feasible implementation, the step of calculating the similarity distance between the basic flavor perception vector of each hot pot base in the basic flavor perception vector library and the target flavor perception vector includes: determining the importance weight of each flavor feature dimension based on the user preference pattern reflected by the user flavor perception vector or the historical user flavor perception vector sequence, or the importance weight information actively set by the user; and using a weighted distance metric, calculating the weighted difference value of the predicted target flavor perception vector and the basic flavor perception vector of each hot pot base in the basic flavor perception vector library on each flavor feature dimension as the similarity distance.

[0094] In this embodiment, importance weight refers to the influence parameter of each flavor feature dimension in similarity calculation, dynamically adjusted according to user preference patterns or actively set information. Specifically, it can be implemented by extracting the frequency of user attention to each dimension from historical vector sequences using statistical analysis methods, or by receiving explicit weight configuration through the user interface, reflecting the differentiated importance that different users attach to features such as spiciness and numbing sensation. The weighted distance measurement method is a mathematical method that combines the difference values ​​of each flavor feature dimension with their corresponding importance weights. For example, it can use the weighted Euclidean distance or weighted Manhattan distance formula. By giving higher weight dimensions a greater weight in the calculation, it ensures that the recommendation results are more in line with the user's core needs.

[0095] In this embodiment, during implementation, the preference distribution pattern in the user's historical flavor perception vector sequence is first analyzed. For example, if a user's quantified value for the spiciness dimension is consistently higher than other dimensions, the importance weight of the spiciness dimension is automatically increased. If the user adjusts the numbness weight to the highest priority through the settings interface, this explicit configuration is directly adopted. Subsequently, for each hot pot base's basic flavor perception vector, the difference value between it and the target vector in each dimension is calculated. For example, the difference in spiciness is 0.8, and the difference in numbness is 0.5. The difference values ​​are multiplied by their corresponding weights and then summed to obtain the total similarity distance. In this way, when a user pays special attention to numbness, bases with larger numbness differences will be assigned higher distance values, thus ranking them lower in the recommendation sort.

[0096] Compared to existing technologies, traditional methods typically use fixed weights or simple arithmetic averages to calculate similarity, failing to dynamically adjust the importance of each flavor dimension based on actual user preferences. For example, in existing technologies, spiciness and numbing sensation are always weighted equally, leading to inaccurate recommendations for users with extreme spiciness preferences. This solution, however, uses a dynamic weight allocation mechanism to accurately capture the differences in user attention to different flavor dimensions. Combined with weighted distance calculation, it significantly improves the accuracy of personalized recommendations. Through this technical solution, this application effectively solves the adaptation bias problem caused by traditional recommendation methods ignoring user-specific weights, achieving accurate matching based on dynamic weights of multi-dimensional flavor features. For example, in scenarios where users highly value the layers of umami, the system can automatically increase the calculated weight of this dimension, prioritizing the recommendation of base ingredients with higher umami feature matching, thereby meeting users' refined needs for specific flavor dimensions.

[0097] In one feasible implementation, the step of generating and outputting a hot pot base recommendation list based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base includes: judging based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base; if the similarity distance between the basic flavor perception vector of all hot pot bases in the basic flavor perception vector library and the predicted target flavor perception vector exceeds a preset mismatch threshold, then based on the target flavor perception vector, a pre-trained regression model is used to deduce the target physicochemical index combination requirements; the regression model is used to characterize the mapping relationship between the flavor perception vector and the key physicochemical indicators of the hot pot base, including capsaicin content, Sichuan pepper content, and free amino acid composition, and the regression model is trained based on historically accumulated sensory evaluation data and physicochemical testing data of hot pot bases; based on the derived target physicochemical index combination requirements, combined with a pre-stored ingredient library and a pre-stored ratio rule library, a custom base recipe suggestion that meets the target physicochemical index combination requirements is generated through an optimization algorithm, and the custom base recipe suggestion is included in the final output hot pot base recommendation list.

[0098] In this embodiment, the preset mismatch threshold refers to a pre-set critical value used to judge the degree of difference between the basic flavor perception vector and the target flavor perception vector. Specifically, it can be determined by using the statistical relationship between user acceptance and flavor difference in historical data, and is used to trigger the custom formula generation mechanism. The regression model refers to a mathematical model established through machine learning that can map the flavor perception vector to a combination of physicochemical indicators. Specifically, it can be a multi-layer neural network trained based on sensory evaluation data and physicochemical test data, used to analyze the correlation between flavor perception and material components. The target physicochemical indicator combination requirement refers to the range constraints of key parameters such as capsaicin content, Sichuan pepper content, and free amino acid composition required to meet the target flavor perception. Specifically, it can be obtained through numerical analysis of the output layer nodes of the regression model, and is used to guide formula optimization. The ingredient library refers to a database storing various base ingredients and their physicochemical properties, specifically including capsaicin content of chili varieties, Sichuan pepper content, and amino acid composition data of bone broth ingredients. The proportioning rule library refers to the set of restrictions and process constraints on the proportion of raw materials in the base ingredient formula, specifically including raw material incompatibilities, maximum addition thresholds, and parameters on the influence of processing temperature on components. Optimization algorithms refer to computational methods for finding combinations of raw materials that meet target physicochemical indicators and conform to proportioning rules. Specifically, genetic algorithms or particle swarm optimization algorithms can be used to search for the optimal formula within the feasible solution space.

[0099] In this embodiment, when the system detects that the difference between the flavor vectors of all pre-stored base ingredients and the target vector exceeds a threshold, it indicates that the existing base ingredients cannot meet the user's needs. At this time, the regression model inversely derives the target range of physicochemical indicators such as capsaicin, ephedrine, and amino acids based on the target flavor vector. The ingredient library provides raw material component data, and the proportioning rule library limits the compatibility of raw materials and process conditions. Under the premise of meeting the target physicochemical indicators, the optimization algorithm traverses the ingredient combinations and proportioning schemes to generate a custom recipe that meets the flavor requirements and is process-feasible. For example, when the user expects a high level of umami flavor but the existing base ingredients lack sufficient umami components, the system may recommend adding a specific proportion of bonito powder and bone broth concentrate, and adjusting the cooking temperature parameters. The final recommendation list includes both standard base ingredients and custom recipe options for the user to choose from.

[0100] In this embodiment, the proposed solution establishes a quantitative relationship between flavor and components through a regression model, and combines this with an optimization algorithm to automatically generate recipes, thus solving the deficiency of existing technologies in covering extremely personalized needs. In this way, when pre-stored base ingredients cannot meet user needs, this application can automatically generate custom recipes that conform to the target flavor characteristics, avoiding the problem of excessive deviation between recommended results and user expectations. By transforming abstract flavor perception into specific physicochemical index constraints and combining them with a raw material database for recipe optimization, the accuracy and feasibility of flavor matching are improved. This mechanism effectively expands the coverage of the recommendation system, enabling users' special flavor needs to be met through systematic calculation, reducing the cost of manual trial and error.

[0101] In one feasible implementation, the method further includes: receiving user feedback data on the hot pot base recommendations in the recommended list; the feedback data includes newly generated comment text data, speech-to-text data, rating data, or purchase behavior data generated by the user for the recommended base; using the feedback data as part of new multimodal user data, combined with the environmental data at the time of obtaining the feedback data, and returning to perform the step of performing natural language processing on the text data and speech-to-text data of the user's comments to obtain identified flavor description words and related intensity words and sentiment words.

[0102] In this embodiment, comment text data refers to descriptive feedback from users regarding the flavor, texture, and experience of the recommended base ingredients in written form. Specifically, natural language processing techniques can be used to extract flavor descriptive words, intensity words, and sentiment words to update the user's flavor preference model. Speech-to-text data refers to feedback expressed by users in speech form, which can be converted into text using speech recognition technology for processing, addressing the issue of diverse user input methods. Rating data refers to users' quantitative evaluations of the recommended base ingredients, which can be converted into weight adjustments for flavor feature dimensions using numerical mapping to calibrate the accuracy of the recommendation model. Purchase behavior data refers to records of whether users actually purchased the recommended base ingredients, which can be obtained using implicit feedback analysis methods to infer users' true satisfaction with the recommendation results from their purchase decisions. Environmental data refers to external conditions such as season and temperature at the time of feedback, which can be obtained by associating timestamps with geographic location information to analyze the dynamic impact of environmental factors on users' flavor preferences.

[0103] In this embodiment, after a user receives a list of recommended hot pot bases and tries them, the system actively collects subsequent multi-dimensional feedback. For example, the user may submit new text reviews through a mobile application or describe their eating experience through a voice assistant. The system converts this unstructured data into structured flavor feature information. Simultaneously, the system records whether the user makes a second purchase of the recommended base or provides a clear evaluation through a five-star rating system. This feedback data, bound to current environmental parameters (e.g., outdoor temperature below 5 degrees Celsius in winter), is re-entered into the data processing flow. By performing natural language processing again on the newly generated review text and voice data, the system can extract updated flavor descriptive words and their associated intensity and sentiment information, thereby adjusting the user's flavor perception vector. Combining historical data and the changing trends of environmental parameters, the system dynamically updates the association weights between user nodes and flavor feature dimension nodes in the heterogeneous graph, ensuring that subsequent recommendation results reflect the user's latest preference changes.

[0104] In some specific implementations, when user feedback includes a voice comment such as "the numbing and spicy flavor lasts better in low-temperature environments," the system can establish a correlation between "low-temperature environment" and the "duration of numbing and spicy flavor," and prioritize bases with high numbing and spicy persistence in subsequent winter scenario recommendations. If a user gives a low rating to a recommended base but completes the purchase, the system can analyze the discrepancy between the purchase behavior and the rating to determine if the user has a potential need to try new flavors, thereby introducing appropriate new category exploration strategies in subsequent recommendations.

[0105] In this embodiment, the solution re-injects feedback data into the data processing flow, enabling the recommendation system to learn online. Simultaneously, by combining the time-series characteristics of environmental parameters, it addresses the problem of static models struggling to adapt to seasonal changes and thus mitigating taste preference shifts. This achieves dynamic optimization capabilities, allowing the flavor matching model to continuously iterate based on real-time user feedback, resolving the recommendation bias issues caused by lagging data updates in traditional methods. Furthermore, by linking feedback data with environmental parameters, the system's adaptability to the influence of external factors such as climate and scene on flavor perception is enhanced, improving the accuracy of recommendation results under different spatiotemporal conditions. Moreover, the comprehensive utilization of multimodal feedback data effectively captures both explicit and implicit user preferences, avoiding the information limitations of single-rate rating data.

[0106] In the embodiments of this application, the big data-based intelligent adaptation method for hot pot base flavors integrates multimodal user data and environmental factors to construct a heterogeneous spectrum, and combines it with a dynamic prediction model to generate accurate flavor recommendations. This effectively solves the shortcomings of traditional methods that cannot analyze user segmentation needs and ignore environmental influences, and can improve the accuracy of hot pot base flavor adaptation to users.

[0107] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the intelligent adaptation method of hot pot base flavor based on big data in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0108] This application also provides a big data-based intelligent flavor adaptation system for hot pot base, see reference. Figure 4 The big data-based intelligent adaptation system for hot pot base flavor includes: a memory 10, a processor 20, and a big data-based intelligent adaptation program for hot pot base flavor stored on the memory 10 and executable on the processor 20. The big data-based intelligent adaptation program for hot pot base flavor is configured to implement the steps of the big data-based intelligent adaptation method for hot pot base flavor.

[0109] The big data-based intelligent adaptation system for hot pot base flavor provided in this application employs the big data-based intelligent adaptation method for hot pot base flavor in the above embodiments, which can improve the accuracy of the adaptation between hot pot base flavor and the user. Compared with the prior art, the beneficial effects of the big data-based intelligent adaptation system for hot pot base flavor provided in this application are the same as those of the big data-based intelligent adaptation method for hot pot base flavor provided in the above embodiments, and other technical features of the big data-based intelligent adaptation system for hot pot base flavor are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0110] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0111] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. All equivalent structural transformations made under the technical concept of this application using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included within the scope of patent protection of this application.

Claims

1. A method for intelligent flavor adaptation of hot pot base based on big data, characterized in that, The method includes: Acquire multimodal user data; the multimodal user data includes text data of user comments, preset taste preference tag data, user historical behavior data, speech-to-text data, and environmental data including season and temperature; Natural language processing is performed on the text data and speech-to-text data of the user comments to obtain the identified flavor description words and related intensity words and sentiment words; Based on the identified flavor descriptive words and related intensity and sentiment words, combined with the preset flavor preference tag data, an analytical model that integrates flavor domain knowledge is used to process the data to obtain the user flavor perception vector; the user flavor perception vector contains quantified values ​​of multiple predefined flavor feature dimensions; Construct a heterogeneous graph containing user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes. Based on the user's historical behavior data, the user's flavor perception vector, the environmental data, and a pre-stored basic flavor perception vector library containing basic flavor perception vectors corresponding to each hot pot base, establish the association relationships between user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes in the heterogeneous graph. Based on the heterogeneous spectrum, the historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data, the target flavor perception vector expected by the user is predicted. Calculate the similarity distance between the basic flavor perception vector of each hot pot base in the basic flavor perception vector library and the target flavor perception vector; Based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base, a hot pot base recommendation list is generated and output. The steps for establishing the association relationships between user nodes, hot pot base nodes, flavor feature dimension nodes, and environmental data nodes in the heterogeneous graph include: Based on the user's historical behavior data, establish a preference relationship edge between the user node and the hot pot base node; Based on the quantized values ​​of each flavor feature dimension in the user's flavor perception vector, a perception intensity relationship edge is established between the user node and the flavor feature dimension node, and the perception intensity relationship edge has a weight representing the quantized value. Based on the specific quantized values ​​of each flavor feature dimension in the basic flavor perception vector of each hot pot base in the basic flavor perception vector library, a strong relationship edge is established between the hot pot base node and the flavor feature dimension node. The strong relationship edge has a weight representing the quantized value of the hot pot base in each flavor feature dimension. Based on environmental data including seasons and temperature, and combined with the changes in user flavor preferences under different environments obtained by analyzing the correlation between user historical behavior data and historical environmental data, an influence relationship edge is established between environmental data nodes and user nodes on the perceived intensity of specific flavor feature dimensions. The influence relationship edge is used to represent the potential influence pattern of the environment on the user's expected changes in flavor perception. The steps for generating and outputting a hot pot base recommendation list based on the similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base include: The similarity distance between the basic flavor perception vector and the target flavor perception vector of each hot pot base is used to determine the similarity distance. If the similarity distance between the basic flavor perception vector of all hot pot bases in the basic flavor perception vector library and the predicted target flavor perception vector exceeds a preset mismatch threshold, then based on the target flavor perception vector, a pre-trained regression model is used to deduce the target physicochemical index combination requirements. The regression model is used to characterize the mapping relationship between the flavor perception vector and the key physicochemical indicators of the hot pot base, such as capsaicin content, Sichuan pepper content, and free amino acid composition. This regression model is trained based on historically accumulated sensory evaluation data and physicochemical test data of hot pot bases. Based on the derived target physicochemical index combination requirements, combined with the pre-stored ingredient library and pre-stored ratio rule library, a custom base sauce recipe suggestion that meets the target physicochemical index combination requirements is generated through an optimization algorithm, and the custom base sauce recipe suggestion is included in the final output hot pot base sauce recommendation list.

2. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 1, characterized in that, Based on the identified flavor descriptors and related intensity and sentiment words, combined with the preset flavor preference tag data, the steps to obtain the user's flavor perception vector using an analytical model that integrates flavor domain knowledge include: By identifying and filtering flavor descriptive words and related intensity and sentiment words that effectively involve predefined flavor feature dimensions through a pre-built ontology library of hot pot flavor domain, effective flavor description data and related effective intensity and sentiment data can be obtained. The effective flavor description data is processed using a pre-trained language model to generate a flavor description semantic vector containing word semantic information. The effective intensity data and effective sentiment data are mapped to intensity quantization values ​​and sentiment quantization values ​​of corresponding predefined flavor feature dimensions; Based on the flavor description semantic vector, intensity quantification value, and sentiment quantification value, combined with preset flavor preference tag data, the user flavor perception vector is generated.

3. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 2, characterized in that, The steps of converting the effective flavor description data into its flavor description semantic vector in the flavor semantic space include: The effective flavor description data is processed using a pre-trained language model as the basic encoder, and higher weights are assigned to words related to predefined flavor feature dimensions during the self-attention calculation process of the pre-trained language model to enhance the representation ability of flavor semantics, thereby outputting the flavor description semantic vector.

4. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 2, characterized in that, The steps of mapping the effective intensity data and effective sentiment data to intensity quantization values ​​and sentiment quantization values ​​of corresponding predefined flavor feature dimensions include: A lightweight classifier is used to identify the intensity level of effective intensity data in a predefined flavor feature dimension, and the identified intensity level is converted into a corresponding numerical quantification value, which is then used as the intensity quantification value. A lightweight classifier is used to identify the emotional polarity category and its intensity level of effective emotional data in a predefined flavor feature dimension, and the identified emotional polarity category and intensity level are converted into corresponding numerical quantification values ​​as the emotional quantification values.

5. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 1, characterized in that, The steps for predicting the user's desired target flavor perception vector based on the heterogeneous spectrum, a historical user flavor perception vector sequence composed of multiple acquired user flavor perception vectors, and the environmental data include: The heterogeneous graph is processed using a graph neural network model. The input includes user nodes, historical user flavor perception vector nodes associated with the user nodes through perception strength relationship edges, hot pot base nodes associated with the user nodes through preference relationship edges, environmental data nodes that the user nodes have historically established associations with, and information about the user's current environment. By propagating and aggregating information along the relationship edges between nodes on the heterogeneous graph, the dynamic influence pattern of environmental changes on user flavor expectations is learned, and the target flavor perception vector is finally output.

6. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 1, characterized in that, The steps for calculating the similarity distance between the basic flavor perception vector of each hot pot base ingredient in the basic flavor perception vector library and the target flavor perception vector include: Based on the user preference patterns reflected by the user flavor perception vector or the historical user flavor perception vector sequence, or the importance weight information actively set by the user, the importance weight of each flavor feature dimension is determined. The weighted distance metric is used to calculate the weighted difference value of the predicted target flavor perception vector and the basic flavor perception vector of each hot pot base in the basic flavor perception vector library in each flavor feature dimension as the similarity distance.

7. The method for intelligent flavor adaptation of hot pot base based on big data as described in claim 1, characterized in that, The method further includes: Receive user feedback data on the hot pot base recommendations in the list; the feedback data includes newly generated comment text data, speech-to-text data, rating data, or purchase behavior data generated by the user for the recommended base they tried; The feedback data is used as part of new multimodal user data. Combined with the environmental data when the feedback data was acquired, the process returns to the step of performing natural language processing on the text data and speech-to-text data of the user comments to obtain the identified flavor description words and related intensity words and sentiment words.

8. A big data-based intelligent flavor adaptation system for hot pot base, characterized in that, The big data-based intelligent adaptation system for hot pot base flavor includes: a memory, a processor, and a big data-based intelligent adaptation program for hot pot base flavor stored in the memory and executable on the processor. The big data-based intelligent adaptation program for hot pot base flavor is configured to implement the steps of the big data-based intelligent adaptation method for hot pot base flavor as described in any one of claims 1 to 7.

Images