Dining recommendation method and system applied to digital multimedia online platform

Abstract

The application provides a catering recommendation method and system applied to a digital multimedia online platform, relates to the technical field of the digital multimedia online platform, and first acquires user catering interaction intention data and platform internal catering content supply time sequence data; constructs a user catering intention evolution model based on the user catering interaction intention data and the platform internal catering content supply time sequence data, forms an intention evolution link; establishes intention supply time sequence adaptation logic in combination with the model and the supply time sequence data; screens content from a platform catering content library according to the adaptation logic to generate a dynamic recommendation scheme; pushes the dynamic recommendation scheme to a user interface and captures new interaction intention data, integrates the original data to update the model and the adaptation logic, so that the dynamic change of the user intention can be accurately grasped, dynamic recommendation is combined with supply time sequence characteristics, and the user consumption experience and platform competitiveness are improved.

Description

Technical Field

[0001] This application relates to the field of digital multimedia online platform technology, and more specifically, to a method and system for restaurant recommendation applied to a digital multimedia online platform. Background Technology

[0002] With the booming development of digital multimedia online platforms, the food and beverage service, as an important business segment, faces increasingly fierce competition. Users' needs for restaurant recommendations are becoming more diverse and personalized, expecting to obtain dining content that accurately matches their own needs and preferences.

[0003] Existing restaurant recommendation methods largely suffer from numerous limitations. Some methods rely solely on users' historical consumption records, neglecting the dynamic changes in user intent. Users' dining needs are not static; they are influenced by factors such as time, context, and mood. Historical data alone cannot accurately capture a user's current needs. Other methods, while considering restaurant supply information, perform only simple static matching without in-depth analysis of the temporal characteristics of supply, failing to provide accurate recommendations based on the availability of food at different stages. Furthermore, existing methods lack effective feedback mechanisms after recommendations, failing to adjust recommendation strategies promptly based on user reactions to new content. This leads to a gradual decline in recommendation effectiveness, failing to meet the ever-changing needs of users and the evolving requirements of digital multimedia online platforms. Summary of the Invention

[0004] In view of this, the purpose of this application is to provide a method and system for restaurant recommendation applied to a digital multimedia online platform.

[0005] According to a first aspect of this application, a method for restaurant recommendation applied to a digital multimedia online platform is provided, the method comprising: The system acquires user interaction intent data on a digital multimedia online platform and supply time-series data of catering content within the platform. The interaction intent data includes the intent tendency, intent change nodes, and intent duration corresponding to user interaction behavior. The supply time-series data of catering content includes the supply cycle stage, supply volume change trend, and supply matching intent tags of catering content. Based on the intent change patterns in the catering interaction intent data and the supply time series data of catering content, a user catering intent evolution model is constructed. The user catering intent evolution model takes the intent nodes of users at different time periods as the core and forms an intent evolution link by connecting them through intent association strength. By combining the user's dining intention evolution model with the supply time series data of dining content, an intention supply time series adaptation logic is established. The intention supply time series adaptation logic is used to associate the user's intention evolution stage with the supply cycle stage of dining content. Based on the intention-supply time sequence adaptation logic, select catering content from the platform's catering content library that is in the matching supply cycle stage and fits the user's current intention evolution stage, and generate a dynamic catering recommendation scheme. The dynamic restaurant recommendation scheme is pushed to the user display interface of the digital multimedia online platform. At the same time, the new restaurant interaction intent data of the user to the dynamic restaurant recommendation scheme is captured and integrated into the original restaurant interaction intent data to update the user's restaurant intent evolution model and intent supply time sequence adaptation logic.

[0006] According to a second aspect of this application, a restaurant recommendation system for a digital multimedia online platform is provided. The restaurant recommendation system for a digital multimedia online platform includes a machine-readable storage medium and a processor. The machine-readable storage medium stores machine-executable instructions. When the processor executes the machine-executable instructions, the restaurant recommendation system for a digital multimedia online platform implements the aforementioned restaurant recommendation method for a digital multimedia online platform.

[0007] According to a third aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, and when the computer-executable instructions are executed, the aforementioned method for recommending restaurants on a digital multimedia online platform is implemented.

[0008] Based on any of the above aspects, the technical effect of this application is as follows: First, user interaction intent data and food content supply time-series data are acquired on the platform. By analyzing the intent change patterns in the interaction intent data and the food content supply time-series data, a user food intent evolution model is constructed. This model centers on user intent nodes at different time periods, connecting them through intent correlation strength to form an intent evolution chain. It can present the evolution process of user food intent, effectively overcoming the shortcomings of traditional methods that cannot capture dynamic changes in user intent. Combining the user food intent evolution model with the food content supply time-series data to establish an intent supply time-series adaptation logic, a precise correlation is achieved between the user intent evolution stage and the food content supply cycle stage. The dynamic food recommendation scheme generated based on this adaptation logic can dynamically adjust according to real-time changes in user intent and food content supply, greatly improving the accuracy and timeliness of recommendations. The recommendation scheme is pushed to the user's display interface, and new user food interaction intent data is captured and integrated into the original data to update the model and adaptation logic. This allows for continuous adaptation to changes in user needs, maintaining high recommendation quality and providing users with more personalized and accurate food recommendation services, significantly improving the user's food consumption experience on the digital multimedia online platform. Attached Figure Description

[0009] Figure 1 A flowchart illustrating the restaurant recommendation method applied to a digital multimedia online platform provided in an embodiment of this application is shown. Figure 2 This illustration shows a schematic diagram of the component structure of a restaurant recommendation system for a digital multimedia online platform, provided in an embodiment of this application, for implementing the above-described restaurant recommendation method for a digital multimedia online platform. Detailed Implementation

[0010] Figure 1 This application provides a schematic flowchart of a restaurant recommendation method and system applied to a digital multimedia online platform, with detailed steps including: Step S110: Obtain user's catering interaction intent data and catering content supply time series data on the digital multimedia online platform. The catering interaction intent data includes the intent tendency, intent change nodes and intent duration corresponding to the user's interaction behavior. The catering content supply time series data includes the supply cycle stage of catering content, supply volume change trend and supply matching intent tags.

[0011] This embodiment uses a scenario where a user browses and selects restaurant content on a digital multimedia online platform as an example to illustrate how to obtain the aforementioned data. First, the digital multimedia online platform records various user interactions in real time, such as browsing different restaurant pages, clicking to view details of a dish, saving specific restaurant information, and rating the dining experience. For these interactions, the platform generates corresponding interaction records, each containing information such as the time of the interaction, the relevant restaurant content identifier, and the interaction type. From these interaction records, the user's intentions can be extracted. For example, if a user browses hot pot restaurants multiple times, it indicates a preference for hot pot; focusing on the price range of dishes during browsing reflects an acceptance of restaurant prices; and checking whether the restaurant offers delivery or private room services demonstrates a demand for restaurant services.

[0012] A change-of-intent node refers to the point in time when a user's intent changes. For example, if a user has been browsing hot pot restaurant content for a week, but suddenly starts browsing barbecue restaurant content extensively, then a certain point in time on that day might be marked as a change-of-intent node. The intent duration period is the time interval between two adjacent change-of-intent nodes, within which the user's intent tends to remain relatively stable.

[0013] Regarding the supply timeline data for food and beverage content, the platform includes information on dishes and promotional activities from various restaurants. The supply cycle is divided based on factors such as the launch time and popularity of the food and beverage content. For example, a newly launched summer limited-edition drink might have a supply cycle that includes a new product launch phase, a peak sales phase, a stable supply phase, and a phase before it's removed from shelves. The supply volume trend records the changes in the quantity of the drink available at different stages. For instance, the supply gradually increases during the new product launch phase, peaks during the peak sales phase, and then gradually decreases as the season changes, entering a stable supply phase. Supply matching intent tags are pre-set based on the characteristics of the food and beverage content to match different user intent preferences. For example, the aforementioned summer limited-edition drink might be tagged with "summer refreshment," "drink," and "affordable" to match users' demand for summer drinks, category preferences, and price acceptance levels.

[0014] Before acquiring user data on their dining interactions, the platform must strictly comply with legal and regulatory requirements. Upon a user's first use of the digital multimedia online platform, a "Privacy Policy" must be displayed via a pop-up window, clearly informing the user of the scope of data collection (only dining interaction data, excluding sensitive personal information such as ID numbers and mobile phone numbers), its purpose (for optimizing recommendation schemes), storage period (interaction data is retained for 180 days, and can be retained long-term after anonymization), and user rights (such as accessing, correcting, and deleting data). Users must manually check "Agree" before using the platform's core functions. Refusing authorization only affects personalized recommendation services and does not affect basic browsing functions. The platform provides a "Privacy Settings" page where users can adjust the scope of data authorization at any time, such as disabling unnecessary data collection items like "Collection Behavior Records" and "Review Content Analysis." After adjustment, the system will immediately stop collecting the corresponding type of data and delete historical data from the past 30 days (except for anonymized statistical data). The scope of data collection is strictly limited through technical means: only user interactions within the restaurant recommendation module (such as browsing the recommendation list and clicking on dish cards) are collected, excluding user operations in system modules such as payment and login; the interaction data only records restaurant content identifiers (such as dish IDs) and is not associated with the user's real identity information, using anonymous device IDs for data aggregation. If a user is detected as a minor (through real-name authentication or age input), an enhanced protection mode is automatically activated: only basic browsing behavior data is collected, and the daily interaction data collection volume does not exceed 50% of the adult threshold; personalized recommendation algorithms based on minors' data are disabled, and age-appropriate restaurant content (such as children's meals and family restaurants) is displayed uniformly. Through these measures, it is ensured that the entire data collection process is within the scope authorized by the user and complies with legal requirements for personal information protection, achieving a balance between data utilization and privacy protection.

[0015] Step S120: Based on the intent change patterns in the catering interaction intent data and the supply time series data of catering content, construct a user catering intent evolution model. The user catering intent evolution model takes the intent nodes of users at different time periods as the core and forms an intent evolution link by connecting them through intent association strength.

[0016] After obtaining the aforementioned data on dining interaction intent and supply time series data, we can begin to construct a user dining intent evolution model. The construction of this model requires in-depth analysis of the patterns of user intent changes and the integration of supply time series data to enrich the model's feature dimensions, enabling the model to accurately reflect the evolution of user intent.

[0017] Step S121: Analyze the catering interaction intent data and extract the intent tendency corresponding to each interaction behavior record. The intent tendency includes the user's preference tendency for catering categories, acceptance tendency for catering prices, and demand tendency for catering services.

[0018] When analyzing restaurant interaction intent data, each interaction record can be analyzed in detail. For example, regarding a user's interaction record of clicking to view a hot pot restaurant, analyzing the restaurant's category information can determine if the user has a preference for hot pot; examining the price range of dishes the user focuses on while staying on the restaurant's page can extract the user's price acceptance tendency; and based on whether the user checks the restaurant's delivery service and whether it has private rooms, it can determine their demand tendency for restaurant services. For user favorites, adding a restaurant that offers 24-hour service to favorites indicates a certain demand tendency for all-day dining services. By analyzing each interaction record in the above way, various user interactions can be transformed into corresponding intent data.

[0019] Step S122: Arrange all interaction behavior records in the order of interaction time to form an intent time series, and mark the time points in the intent time series where the intent tendency changes as intent change nodes.

[0020] All the parsed interaction records are arranged chronologically to form a continuous intent time series. Within this series, changes in user intent can be closely monitored. For example, in the case where a user initially browsed hot pot restaurant content and then suddenly switched to barbecue restaurant content, the initial interaction records in the intent time series were all related to hot pot, corresponding to a hot pot preference. However, from a certain point in time, the interaction records changed to be related to barbecue, indicating a change in intent. This point in time is then marked as the intent change node.

[0021] Step S1221: Using the time interval between two adjacent intention change nodes as an intention duration period, divide the intention time series into multiple consecutive intention duration periods, with each intention duration period corresponding to a stable intention tendency.

[0022] After marking the intent change nodes, the time interval between two adjacent intent change nodes constitutes an intent duration period. For example, if the first intent change node occurs on Monday morning and the second intent change node occurs on Friday evening, then the time interval from Monday morning to Friday evening is an intent duration period. Within this period, the user's intent tendencies are relatively stable, such as a consistent preference for barbecue, an acceptance of mid-to-low price ranges, and a demand for dine-in services. Through this segmentation, the intent time series is divided into multiple continuous intent duration periods with stable intent tendencies.

[0023] Step S1222: Statistically analyze the performance characteristics of intent tendencies within each intent duration period. The performance characteristics include the frequency of occurrence of intent tendencies within the intent duration period, the degree of correlation with interactive behavior, and the degree of influence on subsequent intent changes.

[0024] For each intent duration period, it is necessary to statistically analyze the characteristics of its intent tendency. The frequency of intent tendency refers to the number of times a certain intent tendency occurs within that period. For example, within the intent duration period of the barbecue category preference mentioned above, if a user browses the barbecue restaurant page multiple times, clicks to view barbecue dishes multiple times, and favorites the barbecue restaurant multiple times, then the frequency of the barbecue category preference tendency can be statistically analyzed by the number of these interactive behaviors.

[0025] The degree of correlation between intention and interactive behavior is an indicator that measures the relationship between user intent and interactive behavior. This can be determined by analyzing the types and frequencies of interactive behaviors corresponding to the intention. For example, a user's preference for barbecue products is primarily reflected in browsing and clicking, and these interactions are frequent, indicating a strong correlation between the intention and the interactive behavior.

[0026] The degree of influence on subsequent changes in intent is determined by considering the impact of the current intent tendency on the intent tendency of the next intent duration. If the intent tendency of the current intent duration is related to it in the next intent duration after the current intent duration ends, such as shifting from a preference for barbecue to a preference for grilled skewers, then the current intent tendency has a greater impact on subsequent changes in intent; conversely, if it shifts to a completely unrelated category, the impact is smaller.

[0027] Step S123: Based on the intent tendency and performance characteristics of each intent duration period, construct intent nodes. Each intent node includes a period identifier, core intent tendency, performance characteristic description, and corresponding list of interactive behaviors.

[0028] After analyzing the intent tendency and performance characteristics of each intent duration, intent nodes can be constructed. A duration identifier uniquely identifies the intent duration corresponding to the intent node; for example, the start and end times of a time interval can be used as the identifier. The core intent tendency is the most prevalent intent tendency within the intent duration. For example, in the aforementioned intent duration of barbecue category preference, the core intent tendency is the preference for barbecue category. The performance characteristic description is a textual description of the performance characteristics of the intent node, including the frequency of the intent tendency, the degree of correlation with interactive behavior, and an assessment of its impact on subsequent intent changes. The corresponding interactive behavior list will list all interactive behavior records related to the core intent tendency within the intent duration, facilitating subsequent analysis and querying.

[0029] Step S124: Calculate the intent association strength between two adjacent intent nodes. The intent association strength is obtained by statistically analyzing the degree of overlap of intent tendencies of the two adjacent intent nodes, the smoothness of intent changes, and the continuity of interactive behavior.

[0030] Step S1241: Extract two adjacent intent nodes, and denot them as the preceding intent node and the following intent node, respectively.

[0031] For example, the preceding intent node corresponds to the user's preference for barbecue, while the following intent node corresponds to the user's preference for grilled skewers.

[0032] Step S1242: Extract the core intent tendency from the preceding intent nodes to form a preceding intent tendency set; extract the interaction behavior type and interaction frequency from the interaction behavior list corresponding to the preceding intent nodes to form a preceding interaction behavior set.

[0033] The preceding intent tendency set contains the core intent tendency of the preceding intent node, such as preference for barbecue category, corresponding price acceptance tendency, and service demand tendency. The preceding interaction behavior set contains the interaction behavior type corresponding to the intent node, such as browsing, clicking, and saving, as well as the frequency of each interaction behavior.

[0034] Step S1243: Extract the core intent tendency from the subsequent intent nodes to form a set of subsequent intent tendencies; extract the interaction behavior type and interaction frequency from the interaction behavior list corresponding to the subsequent intent nodes to form a set of subsequent interaction behaviors.

[0035] The set of subsequent intent tendencies includes the core intent tendencies of the subsequent intent nodes, such as preference for barbecue skewers, corresponding price acceptance tendencies, and service demand tendencies. The set of subsequent interaction behaviors also includes the type and frequency of interaction behaviors.

[0036] Step S1244: Count the number of intents with the same type and content in the preceding intent set and the subsequent intent set, and record them as the number of overlapping intents; calculate the proportion of the number of overlapping intents to the total number of preceding intents to obtain the degree of intent overlap.

[0037] For example, the preceding intention set contains three intentions: preference for barbecue, acceptance of low to medium prices, and demand for dine-in services. The subsequent intention set contains three intentions: preference for grilled skewers, acceptance of low to medium prices, and demand for takeout services. The intention of accepting low to medium prices is the same type and has identical content, so the number of overlapping intentions is one. Since the total number of preceding intention sets is three, the intention overlap degree is the number of overlapping intentions divided by the total number of preceding intentions.

[0038] Step S1245: Calculate the time interval between the end time of the intent duration period of the preceding intent node and the start time of the intent duration period of the subsequent intent node, and record it as the node interval duration; determine the intent change smoothing coefficient based on the node interval duration. The shorter the node interval duration, the larger the intent change smoothing coefficient.

[0039] Assuming the intent duration of a preceding intent node ends at a certain point in time, and the intent duration of a subsequent intent node begins at another point in time, then the node interval is the difference between these two points in time. According to preset rules, if the node interval is within a certain range, the intent change smoothing coefficient is set to a higher value; if the node interval is longer, the intent change smoothing coefficient is set to a lower value.

[0040] Step S1246: Count the number of interaction behaviors of the same type in the previous interaction behavior set and the subsequent interaction behavior set, and record them as the number of continued interactions; calculate the proportion of the number of continued interactions to the total number of previous interaction behaviors to obtain the interaction behavior continuity.

[0041] The set of preceding interactive behaviors includes three types: browsing, clicking, and saving. The set of subsequent interactive behaviors includes three types: browsing, clicking, and rating. Browsing and clicking are the same type of interactive behavior, and their number of subsequent interactions is two. The total number of preceding interactive behaviors is three. Therefore, the interaction continuity is the number of subsequent interactions divided by the total number of interactions.

[0042] Step S1247: Set threshold standards for evaluating intent overlap, intent change smoothness coefficient, and interaction behavior continuity; compare intent overlap, intent change smoothness coefficient, and interaction behavior continuity with the corresponding threshold standards to obtain their respective sub-evaluation results; based on the sub-evaluation results, comprehensively determine the intent association strength level between two adjacent intent nodes.

[0043] Thresholds are set for intent overlap, intent change smoothness coefficient, and interaction behavior continuity. The calculated intent overlap is compared to these thresholds; if it exceeds the threshold, the sub-evaluation result is considered good. Similarly, if the intent change smoothness coefficient and interaction behavior continuity both exceed the thresholds, the sub-evaluation result is considered good. Based on the combined results of these three sub-evaluations, the intent association strength between two adjacent intent nodes is determined to be strong.

[0044] Step S125: Using intent association strength as the connection edge, connect all intent nodes in the order of intent duration to form the initial intent evolution link.

[0045] Based on the calculated intent association strength, the intent nodes are connected in chronological order according to their corresponding intent duration. For example, if the first intent node is a preference for hot pot, the second is a preference for barbecue, and the third is a preference for grilled skewers, with medium and strong intent association strengths respectively, then they are connected sequentially with corresponding labeled connection edges to form the initial intent evolution chain.

[0046] Step S126: Extract the supply cycle stage features corresponding to each intent node from the supply time series data of catering content, and add the supply cycle stage features as supplementary information to the corresponding intent node to enrich the feature dimensions of the intent node.

[0047] By analyzing the supply time-series data of food and beverage content, we can identify the supply cycle stage characteristics corresponding to each intent node. For example, for the intent node regarding preference for grilled skewers, the corresponding food and beverage content might be a grilled skewer set meal from a particular grilled skewer shop, currently in a peak sales phase. Supply cycle stage characteristics include the supply volume change trend, user traffic change trend, and platform recommendation resource allocation ratio during this peak sales phase. Adding these characteristics as supplementary information to the intent node enriches its feature dimensions, ensuring that it includes not only intent preferences and performance characteristics but also the corresponding supply cycle stage characteristics.

[0048] Step S127: Traverse all connection edges in the initial intention evolution link, and adjust the intention association strength according to the supplementary supply cycle stage characteristics so that the intention association strength reflects both the intention change pattern and the supply time sequence matching degree.

[0049] Based on the initial intent evolution chain, the intent association strength of the connecting edges is adjusted by incorporating the supply cycle stage characteristics added to the intent nodes. For example, two adjacent intent nodes may have a medium intent association strength initially. However, after adding supply cycle stage characteristics, it is found that the supply cycle stage corresponding to the former intent node and the supply cycle stage corresponding to the latter intent node have a high degree of matching in terms of supply volume change trends and platform recommended resource allocation ratios. Therefore, the intent association strength between them can be appropriately increased so that it not only reflects the intent change pattern but also the matching degree of supply time sequence.

[0050] Step S128: Remove the connection edges and corresponding intent nodes in the initial intent evolution link whose intent association strength level is lower than the set level threshold and which have no subsequent interaction data support, and retain the intent nodes and connection edges whose intent association strength meets the requirements or have subsequent interaction data support to form an intent evolution link.

[0051] A threshold for intent association strength is set. All connecting edges in the initial intent evolution chain are traversed. If the intent association strength level of a connecting edge is lower than the threshold, and the intent node following that edge lacks subsequent interaction data (meaning the user does not generate new interaction behavior after the intent duration corresponding to that intent node ends), then that connecting edge and its corresponding intent node are removed from the chain. Conversely, if the intent association strength level of a connecting edge meets the requirements, or although the level is lower than the threshold, it is supported by subsequent interaction data, then the corresponding intent node and connecting edge are retained, ultimately forming a complete intent evolution chain.

[0052] Step S129: Based on the intent evolution link, add a model update interface, an intent node query interface, and an association strength calculation interface to construct a user catering intent evolution model.

[0053] After establishing the intent evolution chain, corresponding interfaces need to be added to enable the model to perform functions such as updating, querying, and calculating. The model update interface receives new catering interaction intent data to update the intent evolution chain; the intent node query interface allows other modules of the system to query detailed information about specific intent nodes, such as core intent tendencies and performance characteristics; and the association strength calculation interface is used to recalculate the intent association strength between intent nodes when needed. By adding these interfaces, a complete user catering intent evolution model is constructed.

[0054] Step S130: Combine the user's dining intention evolution model with the supply time series data of dining content to establish an intention supply time series adaptation logic. The intention supply time series adaptation logic is used to associate the user's intention evolution stage with the supply cycle stage of dining content.

[0055] After constructing a model of user dining intent evolution, it is necessary to combine it with the time-series data of dining content supply to establish an intent supply time-series adaptation logic, so as to achieve a precise correlation between the user intent evolution stage and the dining content supply cycle stage.

[0056] Step S131: Analyze the supply time series data of catering content and extract the supply cycle stage of each catering content. The supply cycle stage includes the new product launch stage, the hot-selling stage, the stable supply stage, and the delisting preparation stage of catering content.

[0057] Analyzing the supply time-series data of food and beverage content helps determine the supply cycle stage of each item. For example, a newly launched seafood platter experiences a rapid increase in supply during its first week, primarily driven by platform recommendations; this is the "new product launch" stage. After a period, its popularity leads to a peak in supply, with a significant increase in user visits and purchases, marking the "hot-selling" stage. Subsequently, supply and user visits stabilize, indicating a stable supply stage. Finally, due to seasonal changes or ingredient availability, supply gradually decreases, necessitating discontinuation and entering the "removal from shelves" stage. This analysis clarifies the supply cycle stage for each piece of food and beverage content.

[0058] Step S132: Statistically analyze the supply characteristics of catering content in each supply cycle stage. The supply characteristics include the supply volume change curve of catering content, the user access volume change trend, and the platform recommended resource allocation ratio in that supply cycle stage.

[0059] For each supply cycle stage, its supply characteristics are statistically analyzed. The supply quantity change curve can be plotted by recording the daily supply quantity of food and beverage content during that stage. For example, the supply quantity change curve during the new product launch stage may show a gradual upward trend from low to high, while the peak sales stage shows a high-level, stable curve. The user traffic change trend is obtained by statistically analyzing the changes in the number of times users access food and beverage content during that stage over time. For example, during the peak sales stage, user traffic may show a trend of first increasing and then slightly decreasing. The platform recommendation resource allocation ratio refers to the proportion of the food and beverage content in the platform's recommendation slots, homepage displays, and other recommendation resources during that supply cycle stage. The new product launch stage and the peak sales stage usually receive a higher recommendation resource allocation ratio.

[0060] Step S133: Extract the core intent tendency and performance characteristics of each intent node from the user's dining intent evolution model to form an intent feature set, which includes the periodic identifier of the intent node, core intent tendency parameters and performance characteristic parameters.

[0061] From the constructed user dining intent evolution model, the core intent tendency and performance characteristics of each intent node are extracted sequentially. For example, the period of a certain intent node is identified by a specific time interval. The core intent tendency parameters include the degree of preference for seafood, the degree of acceptance of mid-to-high price ranges, and the degree of demand for delivery services; the performance characteristic parameters include intent duration, frequency of interaction behavior, and intent stability indicators. These parameters are integrated to form the intent feature set of that intent node.

[0062] Step S134: Perform correlation analysis between the intent feature set of each intent node and the supply cycle stage characteristics of each catering content to generate intent supply fit. The correlation analysis is achieved by comparing the overlap of intent features and supply features, the degree of matching between demand and supply, and the degree of synchronization in the time dimension.

[0063] The intention feature set of the intention node is correlated with the supply cycle stage characteristics of the catering content. The comparison of overlapping items is to see if there is any overlap between the core intention tendency in the intention feature and the supply matching intention tag in the supply feature. For example, if the intention feature contains "seafood category preference" and the supply matching intention tag in the supply feature also contains "seafood", then there is an overlap.

[0064] The degree of matching between supply and demand is to assess whether the user's intended needs match the supply of catering content. For example, whether the user's acceptance of mid-to-high price range is consistent with the price positioning of catering content, and whether the user's demand for delivery services is consistent with whether catering content provides delivery services.

[0065] The degree of synchronization in the time dimension determines whether the duration of the intent at an intent node is synchronized with the supply cycle of the food and beverage content. For example, if the duration of the intent at an intent node and the peak sales phase of the food and beverage content overlap significantly in time, then the degree of time synchronization is high. By combining the analysis results from these three aspects, the intent supply fit is generated.

[0066] Step S135: Based on the intent supply adaptation degree, assign a corresponding supply cycle stage matching label to each intent node. The supply cycle stage matching label includes the primary adaptation supply stage, the secondary adaptation supply stage, and the adaptation priority.

[0067] Based on the generated intent supply fit, supply cycle stage matching tags are assigned to intent nodes. If an intent node has the highest fit with the hot-selling stage of food and beverage content, then the primary fit supply stage is marked as the hot-selling stage; the fit with the stable supply stage is the next highest, and it is marked as the secondary fit supply stage. The fit priority is then ranked according to the fit degree, with the hot-selling stage having the highest priority and the stable supply stage having the second highest priority.

[0068] Step S136: Count the matching frequency of all intent nodes with each supply cycle stage, and generate an intent supply matching frequency distribution table. The intent supply matching frequency distribution table records the total number of times each supply cycle stage is matched, the type distribution of the corresponding intent nodes, and the average fit.

[0069] The matching results of all intent nodes with each stage of the supply cycle were statistically analyzed. For example, the statistics showed that the new product launch stage was matched several times, with the corresponding intent node types mainly being those who preferred new dishes, and the average fit was a certain value; the hot-selling stage was matched several times, with the corresponding intent node types being diverse, and the average fit was also a certain value, etc. The above statistical results were compiled into an intent supply matching frequency distribution table.

[0070] Step S137: Based on the intention supply matching frequency distribution table, determine the adaptation rules for each intention node and the corresponding supply cycle stage. The adaptation rules include the adaptation degree calculation standard, the supply cycle stage matching label allocation conditions, and the adaptation priority ranking basis.

[0071] Based on the data in the intention-to-supply matching frequency distribution table, the matching rules are determined. The matching degree calculation standard can be adjusted according to the average matching degree of different supply cycle stages. For example, for the peak sales stage, since its average matching degree is higher, the weight of the demand-supply matching degree in the matching degree calculation can be appropriately increased. The supply cycle stage matching label allocation condition stipulates that when the intention-to-supply matching degree reaches a certain threshold, the corresponding supply cycle stage is marked as a primary matching supply stage or a secondary matching supply stage. The matching priority ranking is explicitly based on the intention-to-supply matching degree from high to low. When the matching degrees are the same, the matching frequency of the supply cycle stage is taken into account.

[0072] Step S138: Collect historical user intent evolution data and corresponding supply time series data, verify the effectiveness of the adaptation rules using historical user intent evolution data and corresponding supply time series data, and adjust the adaptation rule parameters by comparing the degree of fit between the adaptation results output by the adaptation rules and the actual historical interaction effects.

[0073] Step S1381: Extract the catering interaction intent data of multiple historical users from the historical data storage module of the digital multimedia online platform to form a historical intent evolution dataset. The historical intent evolution dataset includes the intent node sequence, intent association strength and intent change pattern of each historical user.

[0074] From the platform's historical data storage module, we filter out the dining interaction intent data of multiple users over a period of time. This data includes each user's intent node sequence, i.e., the intent nodes of a user at different times arranged in order; intent association strength, i.e., the association strength level between each intent node; and intent change patterns, such as the change pattern and frequency of intent tendency. Integrating this data forms a historical intent evolution dataset.

[0075] Step S1382: Extract the time series data of catering content supply corresponding to the historical intent evolution dataset to form a historical supply time series dataset. The historical supply time series dataset includes the supply cycle stage, supply characteristics and supply matching intent labels of catering content in each historical time period.

[0076] Based on the time range in the historical intent evolution dataset, the time series data of food and beverage content supply within the corresponding time period is extracted. The above data includes the supply cycle stage, supply characteristics, and supply matching intent tags of each food and beverage content within the historical time period, forming a historical supply time series dataset.

[0077] Step S1383: Input a single historical intent evolution data point from the historical intent evolution dataset into the basic framework of intent supply time-series adaptation logic, and output the corresponding supply cycle stage matching result and intent supply adaptation degree according to the adaptation rules to form a prediction adaptation result.

[0078] A single historical intent evolution data point is selected from the historical intent evolution dataset and input into the basic framework of intent supply time-series adaptation logic. Based on the adaptation rules, this framework analyzes the user's intent node sequence, intent association strength, and other information, and combines this with historical supply time-series data to output the corresponding supply cycle stage matching result and intent supply adaptation degree, thus forming a predicted adaptation result.

[0079] Step S1384: Extract the actual user interaction record corresponding to the historical intent evolution data from the platform's historical data. The actual user interaction record includes the food and beverage content actually selected by the user, the interaction time, and the interaction frequency.

[0080] Search the platform's historical data for actual user interaction records corresponding to the historical intent evolution data. For example, the user actually selected dishes from certain restaurants during a specific time period, the interaction time was concentrated in certain periods, and the interaction frequency was a specific number of times.

[0081] Step S1385: Count the number of supply cycle stages that are successfully matched with the actual catering content selected by the user in the statistical prediction and adaptation results, and record them as the number of successful matches. Count the number of all recommended supply cycle stages in the statistical prediction and adaptation results, and record them as the total number of predictions. Calculate the proportion of the number of successful matches to the total number of predictions to obtain the adaptation success rate of a single historical intent evolution data.

[0082] Assuming that the prediction and adaptation results recommend several supply cycle stages, and some of these stages match the supply cycle stage corresponding to the actual food and beverage content selected by the user, then the number of successful matches is this number, the total number of predictions is the total number of recommendations, and the adaptation success rate of a single historical intent evolution data point is the number of successful matches divided by the total number of predictions.

[0083] Step S1386: Repeat the above steps to calculate the adaptation success rate of all historical intent evolution data in the historical intent evolution dataset, and take the average of the adaptation success rates of all historical intent evolution data to obtain the overall adaptation success rate.

[0084] Following the method described above, calculate the adaptation success rate for each data point in the historical intent evolution dataset, then sum all the adaptation success rates and divide by the total number of data points to obtain the overall adaptation success rate.

[0085] Step S1387: If the overall adaptation success rate reaches the adaptation success rate threshold, determine that the current parameters of the adaptation rule are valid; if the overall adaptation success rate does not reach the adaptation success rate threshold, analyze the differences between the predicted adaptation results and the actual user interaction records. The differences include the intention supply adaptation degree calculation deviation, the supply cycle stage matching tag allocation deviation, and the adaptation priority sorting deviation.

[0086] A success rate threshold is set. If the calculated overall success rate reaches the threshold, the current parameters of the adaptation rule are deemed valid; otherwise, discrepancies are analyzed. For example, it may be found that the intended supply adaptation degree is overcalculated in some predicted adaptation results, while actual users do not select catering content for that supply cycle stage. This is an intention supply adaptation degree calculation deviation. Or, the supply cycle stage matching tags may be incorrectly assigned, marking the main adapted supply stage as the new product launch stage, while users actually prefer the best-selling stage. This is a supply cycle stage matching tag assignment deviation. Adaptation priority ranking deviation refers to the recommended priority order of supply cycle stages not matching the user's actual selection order.

[0087] Step S1388: Adjust the adaptation degree calculation standard parameters, supply cycle stage matching label allocation condition parameters, and adaptation priority sorting basis parameters in the adaptation rules according to the differences. After adjustment, re-enter the historical intent evolution data and corresponding supply time series data for verification until the overall adaptation success rate reaches the adaptation success rate threshold, and complete the adjustment of the adaptation rule parameters.

[0088] For the discrepancies identified in the analysis, the adaptation rule parameters are adjusted. For example, if there is a deviation in the calculation of intent-supply fit, it may be due to an unreasonable weighting of the degree of matching between demand and supply; this weighting should be appropriately reduced. If there is a deviation in the allocation of matching tags for different supply cycle stages, the fit threshold for the main supply stages should be adjusted to raise the judgment standard. If there is a deviation in the adaptation priority ranking, the ranking criteria should be modified to increase the influence of the frequency of user historical interactions in the ranking. After the adjustments are completed, the historical intent evolution data and corresponding supply time-series data are re-inputted for verification, and the overall adaptation success rate is calculated. If the threshold is still not reached, the parameters are adjusted again until the overall adaptation success rate reaches the threshold, thus completing the adjustment of the adaptation rule parameters.

[0089] Step S139: Integrate the verified and adjusted adaptation rules, the intent supply adaptation degree calculation method, and the supply cycle stage matching tag allocation mechanism to form the basic framework of intent supply time sequence adaptation logic; add an adaptation logic update module to the basic framework of intent supply time sequence adaptation logic. This adaptation logic update module is used to receive new interactive intent data and supply time sequence data, dynamically adjust the adaptation rules and adaptation degree calculation parameters, and complete the establishment of intent supply time sequence adaptation logic.

[0090] The validated and adjusted adaptation rules, the intent supply adaptation degree calculation method, and the supply cycle stage matching tag allocation mechanism are integrated to form the basic framework of intent supply time-series adaptation logic. Then, an adaptation logic update module is added. This module can receive newly generated interaction intent data and supply time-series data in real time, and dynamically adjust the parameters in the adaptation rules and adaptation degree calculation parameters based on this new data. This ensures that the intent supply time-series adaptation logic can be continuously optimized as the data changes, thereby completing the establishment of the intent supply time-series adaptation logic.

[0091] Step S140: Based on the intent supply time sequence adaptation logic, select catering content from the platform's catering content library that is in the matching supply cycle stage and matches the user's current intent evolution stage, and generate a dynamic catering recommendation scheme.

[0092] After establishing the logic for matching the timing of intent supply, suitable catering content can be selected from the platform's catering content library based on this logic to generate dynamic catering recommendation schemes for users.

[0093] Step S141: Extract the latest intent node from the user's dining intent evolution model as the current intent node, and determine the core intent tendency and main adaptation supply stage corresponding to the current intent node.

[0094] The user's dining intent evolution model updates user intent nodes in real time, extracting the most recently formed intent node as the current intent node. For example, the current intent node is the user's most recently formed preference intent node for Japanese cuisine. Its core intent tendencies include a preference for Japanese cuisine, an acceptance of mid-to-high price ranges, and a demand for higher service standards in the dining environment. According to the intent supply time sequence adaptation logic, the main adaptation supply stage for this current intent node is the hot-selling stage.

[0095] Step S142: Invoke the intent supply timing adaptation logic, query the supply cycle stage characteristics that match the core intent tendency of the current intent node, and obtain the target supply cycle stage characteristics.

[0096] The logic for adapting the timing of intent supply is invoked, inputting the core intent tendency of the current intent node and querying the supply cycle stage characteristics that match it. For example, if the core intent tendency is a preference for Japanese cuisine, the logic will query the supply cycle stage characteristics of all Japanese cuisine content in the platform's food and beverage content library, filtering out those that are in a hot-selling phase and whose supply characteristics match the core intent tendency, such as sufficient supply, high user traffic, and a reasonable allocation of platform recommendation resources. These characteristics together constitute the target supply cycle stage characteristics.

[0097] Step S143: Extract all catering content that is in the target supply cycle stage from the platform's catering content library to form an initial catering content set.

[0098] Based on the characteristics of the target supply cycle stage, all food and beverage content that falls within that target supply cycle stage is selected from the platform's food and beverage content library. For example, if the target supply cycle stage is a peak sales period for Japanese cuisine, then all Japanese cuisine content in this peak sales period is extracted, such as Japanese sushi sets, Japanese BBQ sets, and Japanese ramen sets, forming an initial set of food and beverage content.

[0099] Step S144: Extract the supply characteristics and supply matching intent tags of each catering content in the initial catering content set, and compare the supply matching intent tags with the core intent tendency of the current intent node.

[0100] For each piece of food and beverage content in the initial set of food and beverage content, its supply characteristics are extracted, such as the trend of supply volume changes and the trend of user visits, as well as supply matching intent tags, such as "Japanese cuisine," "mid-to-high price range," and "elegant environment." Then, the above supply matching intent tags are compared one by one with the core intent tendency of the current intent node.

[0101] Step S145: Retain catering content whose supply matching intent tags overlap with the core intent tendency of the current intent node, forming an intermediate catering content set.

[0102] During the comparison process, catering content whose supply matching intent tags overlap with the core intent tendency is retained. For example, a Japanese sushi set meal's supply matching intent tags include "Japanese cuisine," "mid-to-high price," and "elegant environment," which completely overlap with the core intent tendency of the current intent node, so this set meal is retained; a Japanese ramen set meal's tags include "Japanese cuisine," "mid-to-low price," and "fast service," with "Japanese cuisine" overlapping with the core intent tendency, so this set meal is also retained. Through the above filtering, an intermediate catering content set is formed.

[0103] Step S146: Extract the list of catering content that the user has interacted with from the catering interaction intent data, remove catering content from the intermediate catering content set that exceeds the preset duplication threshold in terms of core intent matching degree with the catering content that the user has interacted with, and obtain the final catering content set.

[0104] The system retrieves a list of previously interacted restaurant content from the user's restaurant interaction intent data, such as a previously purchased Japanese sushi set meal. It calculates the core intent match between each restaurant content in the intermediate set and the previously interacted restaurant content, setting a specific duplicate threshold. If another restaurant's Japanese sushi set meal in the intermediate set has a core intent match exceeding the threshold with a previously purchased sushi set meal, it is removed from the intermediate set to avoid duplicate recommendations. This process yields the final set of restaurant content.

[0105] Step S147: Invoke the intent supply timing adaptation logic to calculate the intent supply adaptation degree between each catering content in the final catering content set and the current intent node.

[0106] Step S1471: Extract a single piece of catering content from the final catering content set, obtain the supply time series data corresponding to the catering content, and parse out the supply cycle stage characteristic parameters. The supply cycle stage characteristic parameters include the peak time of supply volume, the growth rate of user visits, and the proportion of platform recommended resources.

[0107] Taking the Japanese BBQ set meal in the final catering content collection as an example, we obtained its supply time series data and analyzed the characteristic parameters of the supply cycle stage. The peak supply time refers to the point in time when the supply of this set meal reaches its maximum during the hot-selling phase; the user visit growth rate refers to the speed at which user visits increase over time during the hot-selling phase; and the proportion of platform recommended resources refers to the proportion of this set meal in the platform's recommended resources.

[0108] Step S1472: Convert the characteristic parameters of the supply cycle stage into a standardized supply characteristic vector.

[0109] The extracted supply cycle stage characteristic parameters are standardized to eliminate the influence of different dimensions. For example, the peak supply time can be converted into a relative time proportion within the peak sales period, and the user traffic growth rate and the proportion of platform recommended resources are also converted into values ​​within specific ranges through corresponding standardization methods, ultimately forming a standardized supply characteristic vector.

[0110] Step S1473: Extract core intent tendency parameters and performance characteristic parameters from the current intent node. The core intent tendency parameters include category preference ratio, price acceptance range ratio, and service demand ratio. The performance characteristic parameters include intent duration, interaction frequency, and intent stability index.

[0111] Relevant parameters are extracted from the current intent node. Category preference percentage represents the proportion of a user's preference for Japanese cuisine among all category preferences. For example, the proportion of a user's interactions with Japanese cuisine out of all interactions with other food and beverage categories in the past month. Price acceptance range percentage is the proportion of a user's acceptable mid-to-high price range out of all their acceptable price ranges. For example, if a user's acceptable price range includes low, medium, and high levels, the proportion of mid-to-high price range is the proportion of interactions with mid-to-high price ranges out of all interactions with other price ranges. Service demand percentage is the proportion of service demands with high requirements for the dining environment out of all service demands. For example, during the interaction process, among service demands such as dining environment, delivery service, and promotional activities, the proportion of demand for the dining environment is considered. Among the performance characteristic parameters, the intent duration is the length of the intent duration period corresponding to the current intent node, that is, the time span from the start of the intent node to the present; the interaction frequency is the number of times user interaction occurs within the period, such as the total number of times browsing, clicking, collecting, and rating; the intent stability index is used to measure the stability of intent tendency within the period, which is determined by calculating the fluctuation range of intent tendency within the period. The smaller the fluctuation range, the higher the stability index.

[0112] Step S1474: Convert the core intent tendency parameters and performance feature parameters into a standardized intent feature vector.

[0113] Similarly, the core intent preference parameters and performance characteristic parameters are standardized. Parameters such as category preference ratio, price acceptance range ratio, service demand ratio, intent duration, interaction frequency, and intent stability index are converted into standardized values ​​corresponding to the dimensions of the standardized supply feature vector. For example, various ratio parameters are normalized to values ​​between 0 and 1; intent duration is compared with the platform's preset longest intent duration to obtain a relative proportion; interaction frequency is standardized by comparing it with the user's historical average interaction frequency; and intent stability index is converted to values ​​between 0 and 1 based on the magnitude of the fluctuation range. This ultimately forms a standardized intent feature vector, where each element corresponds to a standardized feature parameter and is one-to-one with the dimensions of the standardized supply feature vector for subsequent matching degree calculations.

[0114] Step S1475: Call the adaptation degree calculation module in the intent supply timing adaptation logic, and obtain the pre-stored feature weight allocation table in the adaptation degree calculation module. The feature weight allocation table contains weight coefficients corresponding to different feature dimensions, and the weight coefficients are dynamically adjusted according to the historical adaptation effect.

[0115] The adaptation calculation module in the intent supply time-series adaptation logic is invoked. This module pre-stores a feature weight allocation table. The feature weight allocation table assigns a corresponding weight coefficient to each feature dimension of the standardized supply feature vector and the standardized intent feature vector. These weight coefficients reflect the importance of different feature dimensions in the adaptation calculation. For example, the weight coefficient for the platform recommendation resource proportion dimension in the supply features may be high because the allocation of platform recommendation resources often reflects the popularity of food and beverage content; similarly, the weight coefficient for the category preference proportion dimension in the intent features may be high because category preference directly reflects the user's core intent. The weight coefficients are not fixed but dynamically adjusted based on historical adaptation results. For instance, if a certain feature dimension has a significant and accurate impact on the results in multiple adaptations, its weight coefficient can be appropriately increased; conversely, if a certain feature dimension has a small impact or frequently causes adaptation deviations, its weight coefficient can be decreased.

[0116] Step S1476: For each pair of corresponding dimension features in the standardized supply feature vector and the standardized intent feature vector, calculate its matching score; according to the weight coefficient of the corresponding dimension in the feature weight allocation table, calculate the weighted average of the matching scores of all dimensions to obtain the intent supply fit degree between the catering content and the current intent node.

[0117] For each pair of corresponding dimensional features in the standardized supply feature vector and the standardized intent feature vector, such as the "user visit growth rate" dimension in the supply features and the "category preference ratio" dimension in the intent features, a matching score is calculated. The matching score is calculated as follows: First, calculate the absolute difference between the two standardized feature values, then subtract this absolute difference from 1 (when the difference is 0, the matching score is 1; when the difference is 1, the matching score is 0) to obtain the original matching score for that dimension. Next, based on the weight coefficients of the corresponding dimensions in the feature weight allocation table, multiply the original matching score of each dimension by the corresponding weight coefficient to obtain the weighted matching score. Finally, sum the weighted matching scores of all dimensions and divide by the sum of all weight coefficients to obtain the weighted average matching score, which is the suitability of the catering content for the current intent node's intent supply. For example, the original matching degrees of the standardized supply feature vector of a Japanese barbecue set meal and the standardized intent feature vector of the current intent node in the five dimensions are 0.8, 0.7, 0.9, 0.6, and 0.85, respectively, with corresponding weight coefficients of 0.2, 0.15, 0.25, 0.1, and 0.3. The weighted matching degrees are 0.16, 0.105, 0.225, 0.06, and 0.255, respectively, with a total of 0.805. The sum of the weight coefficients is 1, so the intent supply fit is 0.805.

[0118] Step S1477: Repeat the above steps to calculate the intent supply fit degree of all catering content in the final catering content set with the current intent node, associate and store the intent supply fit degree of all catering content with the corresponding catering content identifier, and form a fit degree statistics list.

[0119] For example, step S1477-1: Create a blank adaptation statistics list, which includes a catering content identifier column, an intention supply adaptation column, and a supply cycle stage column.

[0120] Create a blank compatibility statistics list. This compatibility statistics list adopts a structured data format and contains three columns: the catering content identifier column is used to store the unique code of each catering content in the platform for subsequent query and association; the intent supply compatibility column is used to store the calculated compatibility value between the catering content and the current intent node; and the supply cycle stage column is used to store the current supply cycle stage of the catering content, such as "hot selling stage" or "stable supply stage".

[0121] Step S1477-2: Extract the identifier of a single catering content from the final catering content set and fill it into the catering content identifier column of the fit statistics list; extract the intent supply fit corresponding to the catering content and fill it into the intent supply fit column of the fit statistics list; extract the supply cycle stage information of the catering content and fill it into the supply cycle stage column of the fit statistics list.

[0122] Process each food item in the final food item set sequentially. For example, first process the Japanese BBQ set meal, which is tagged "RC-JP-001", has an intended supply fit of 0.805, and is in the "hot-selling phase" of the supply cycle. Fill the corresponding columns in the fit statistics list with this information. Next process the Japanese sushi set meal, which is tagged "RC-JP-002", has a fit of 0.78, and is also in the "hot-selling phase" of the supply cycle. Fill the list with this information as well, and so on, until all food items have been processed.

[0123] Step S1477-3: Repeat the above steps to fill the information of all catering content in the final catering content set into the fit statistics list.

[0124] Following step S1477-2, iterate through each food and beverage item in the final food and beverage content set, and fill in its identifier, intended supply fit, and supply cycle stage information into the fit statistics list one by one, ensuring that each food and beverage item has a corresponding record and that the information is accurate. For example, if the final food and beverage content set contains 5 food and beverage items, the fit statistics list will have 5 rows of records, with each row corresponding to the information of one food and beverage item.

[0125] Step S1477-4: Sort the compatibility statistics list in descending order of the values ​​in the intended supply compatibility column. If the values ​​in the intended supply compatibility column of two catering items are the same, sort them in the order of new product launch stage, hot-selling stage, stable supply stage, and delisting preparation stage of the supply cycle stage column. If the information in the supply cycle stage column of two catering items is also the same, sort them in the order of the coding of the catering item identifier. After sorting, extract the information of the first few catering items in the compatibility statistics list to form an ordered catering item information set.

[0126] First, the matching statistics list is sorted in descending order by the intended supply matching score column, with the highest matching score value placed first, and so on down. When two food and beverage items have the same matching score value, such as a Japanese BBQ set meal and a Japanese sushi set meal both having a matching score of 0.805, they are sorted according to the priority of the supply cycle stage, in the order of new product launch stage > hot-selling stage > stable supply stage > delisting preparation stage. If both are in the hot-selling stage, they are further sorted according to the coding order of the food and beverage content identifiers. The coding order is usually arranged according to the order in which the food and beverage content was entered into the platform's database or the natural order of letters and numbers, for example, "RC-JP-001" is placed before "RC-JP-002". After sorting, according to the platform's preset recommendation quantity (e.g., recommending 5 food and beverage items), the information of the first 5 food and beverage items is extracted from the sorted list, including the food and beverage content identifier, intended supply matching score, and supply cycle stage, forming an ordered food and beverage content information set. The order of the food and beverage items in this ordered food and beverage content information set is the preliminary order of recommendation display.

[0127] Step S1477-5: Associate the catering content identifiers in the ordered catering content information set with the catering content display information in the platform's catering content library to obtain the corresponding catering content images, catering details descriptions, and intent matching instructions; integrate the catering content display information according to the sorting order of the ordered catering content information set to form the core content of the dynamic catering recommendation scheme, and complete the construction of the dynamic catering recommendation scheme.

[0128] Based on the restaurant content identifiers in the ordered restaurant content information set, the platform retrieves the display information corresponding to each restaurant content through its query interface. Restaurant content images include actual photos of dishes and images of the restaurant environment, and these images must conform to the platform's specified size and format. Restaurant details include the dish name, ingredient composition, flavor characteristics, price, promotional information, restaurant address, and opening hours. The intent matching description is a textual description generated based on the matching of the restaurant content's supply matching intent tag with the core intent tendency of the current intent node. For example, "This Japanese BBQ set meal is currently a bestseller, matching your preference for Japanese cuisine and your acceptance of mid-to-high price range. The restaurant environment is elegant, meeting your needs for a pleasant dining environment." The above display information is integrated according to the sorting order of the ordered restaurant content information set. For example, the first item displays the image, details, and matching description of the Japanese BBQ set meal, the second item displays relevant information for the Japanese sushi set meal, and so on, forming the core content of the dynamic restaurant recommendation scheme. The core content also includes metadata such as the unique identifier of the recommendation scheme and the generation timestamp, thus completing the construction of the dynamic restaurant recommendation scheme.

[0129] Step S148: Sort the catering contents in the final catering content set in descending order of intended supply suitability to form an ordered catering content list.

[0130] After obtaining and sorting the matching statistics list, the food and beverage content in the final set is arranged in descending order of its matching score to the intended supply, forming an ordered list of food and beverage content. This ordered list only contains the food and beverage content identifier and its corresponding matching score value, used to clarify the recommendation priority of the food and beverage content. For example, in the sorted list, the first item is a Japanese BBQ set meal with a matching score of 0.85 (identified by RC-JP-001), the second item is a Japanese sushi set meal with a matching score of 0.82 (identified by RC-JP-002), the third item is a Japanese ramen set meal with a matching score of 0.79 (identified by RC-JP-003), and so on. The order in the list directly determines the display order of subsequent recommended content.

[0131] Step S149: Select the first few catering items from the ordered catering content list, extract the display information for each catering item, the display information includes catering content images, catering details descriptions, supply cycle stage identifiers, and intent matching instructions; integrate the selected catering content display information according to the sorting order of the ordered catering content list, add the recommendation generation time, current intent node identifier, and target supply cycle stage identifier to form a dynamic catering recommendation scheme; add scheme update trigger conditions to the dynamic catering recommendation scheme, the scheme update trigger conditions include the occurrence of new user interaction behavior, changes in the current intent node, and updates to the supply cycle stage.

[0132] Select the top few restaurant listings from the ordered restaurant content list. The number of selections is determined based on the platform's recommendation slot capacity and user experience, for example, selecting the top 5. For each selected restaurant listing, extract its complete display information from the platform's restaurant content library, including restaurant images (high-resolution food photos, storefront photos), detailed restaurant descriptions (including dish specifications, cooking methods, user review summaries, store features, etc.), supply cycle stage indicators (such as a text label and corresponding icon for "Hot Selling Stage"), and intent matching instructions (detailing how the restaurant content matches the user's current intent, such as "You currently prefer Japanese cuisine. This set meal uses carefully selected high-quality ingredients, is currently hot selling, is priced within your mid-to-high price range, and provides an elegant dining environment"). Integrate the above display information according to the sorted order of the ordered restaurant content list to form the main body of the recommended content.

[0133] Next, add the recommendation generation time, accurate to the second, to record the moment the solution was generated; the current intent node identifier, which is the unique number of the current intent node in the user's dining intent evolution model, for subsequent tracking and updates; and the target supply cycle stage identifier, such as the code for "hot-selling stage," to clarify the supply cycle stage matched by this recommendation. Combine the above metadata with the main recommendation content to form a complete dynamic dining recommendation solution.

[0134] Finally, update trigger conditions are added to the restaurant dynamic recommendation scheme. When a user engages in new interaction in the recommended content area (such as clicking to view a recommended item, saving it, or rating it), the scheme is updated to reflect the user's latest intention changes. When the current intention node in the user's restaurant intention evolution model changes (such as a user switching from a preference for Japanese cuisine to a preference for Korean cuisine), the scheme is automatically updated. When the supply cycle stage of the recommended restaurant content is updated (such as a set meal moving from the "hot-selling stage" to the "stable supply stage"), the scheme is also updated to ensure that the recommended content remains synchronized with the supply timeline.

[0135] Step S150: Push the dynamic restaurant recommendation scheme to the user display interface of the digital multimedia online platform, and at the same time capture the user's new restaurant interaction intent data for the dynamic restaurant recommendation scheme. Integrate the new restaurant interaction intent data into the original restaurant interaction intent data to update the user's restaurant intent evolution model and intent supply time sequence adaptation logic.

[0136] Step S151: Call the data transmission interface of the digital multimedia online platform to transmit the catering content display information, supply cycle stage identifier and intent matching description in the catering dynamic recommendation scheme to the data transmission interface of the display interface.

[0137] The platform's internal service call mechanism invokes the data transmission interface of the digital multimedia online platform's display interface. This interface adopts a RESTful API architecture and supports data transmission in JSON format. The data, including restaurant content display information (image URLs, detailed description text), supply cycle stage identifiers (text and icon encoding), and intent matching description text from the restaurant dynamic recommendation scheme, is encapsulated into data packets according to the interface's required field format. For example: { "recommendationId":"REC-20240520-001", "generateTime":"2024-05-20T18:30:00Z", "currentIntentNodeId":"INT-005", "targetSupplyStage":"HOT", "contents":[ { "contentId":"RC-JP-001", "imageUrl":"https: / / example.com / images / yakiniku.jpg", "description":"Japanese BBQ set meal for two, including selected short ribs, marbled beef, etc...", "supplyStage":"HOT", "Matching Note": "This matches your preference for Japanese cuisine and your mid-to-high price range..." }, ... ] } The encapsulated data packets are sent to the data transmission interface of the display interface via the HTTPS protocol to ensure the security and integrity of data transmission.

[0138] Step S152: After receiving the data, the display interface data transmission interface organizes the data according to the preset recommended content layout rules of the digital multimedia online platform to generate display interface rendering data. The recommended content layout rules include the display order of catering content, information layout format and interactive entry settings.

[0139] After receiving data packets, the data transmission interface of the display interface parses and verifies the data to ensure that the data format is correct and the content is complete. Then, the data is organized according to the preset recommended content layout rules of the digital multimedia online platform. The layout rules include: the display order of catering content strictly follows the sorting order of the ordered catering content information set, with the first item placed at the top of the recommended area; the information layout format stipulates that catering content images occupy 40% of the width of the display card, with a fixed height, and the detailed description is located to the right of the image, occupying 60% of the width, including the title (bold font, large font size), introduction (medium font size, limited to 2 lines), price (red font), and intent matching description (small gray font); the interactive entry is set at the bottom of the display card, including a "View Details" button (blue background, white text) and an "Add to Favorites" icon button (hollow star icon), with button size and spacing uniformly set according to the platform design specifications. Based on these rules, the raw data is converted into HTML / CSS / JS code snippets or native interface element description data that can be directly rendered by the display interface, generating display interface rendering data.

[0140] Step S153: Send the rendering data of the display interface to the digital multimedia online platform display interface on the user's end, and display the catering content in the recommended content area of ​​the digital multimedia online platform display interface according to the sorting order of the catering dynamic recommendation scheme.

[0141] The generated rendering data is pushed to the user's digital multimedia online platform display interface in real time via the WebSocket protocol, or returned via an HTTP request when the user refreshes the interface. After receiving the rendering data, the user-side display interface is parsed and executed by the front-end rendering engine (such as a browser's rendering engine or the mobile application's native rendering framework). The rendered data is then displayed in the recommended content area (usually located in the middle of the homepage or under the "Recommended for You" section) according to the order of the dynamic restaurant recommendation scheme. The display effects include: each restaurant item is presented in a card-style layout, containing images, text information, and interactive buttons; fixed spacing and separators between cards; the page supports vertical scrolling to browse more recommended content; when the user clicks the "View Details" button, they are redirected to the details page of that restaurant item; when the user clicks the "Add to Favorites" button, the icon changes to a solid star, indicating successful addition to the favorites, and feedback is sent to the server in real time.

[0142] Step S154: Activate the user interaction behavior capture module of the digital multimedia online platform to monitor the user's operation behavior in the recommended content area in real time. The operation behavior includes browsing food and beverage content, clicking on food and beverage content, collecting food and beverage content, and rating food and beverage content.

[0143] While showcasing the dynamic restaurant recommendation solution, a user interaction behavior capture module is activated on the digital multimedia online platform. This module, integrated into the user-side application, captures user behavior by listening to events on page elements (such as mouse hover, clicks, and touches). Specific monitored behaviors include: browsing restaurant content (determined by monitoring user swipes and dwell time in the recommended area; a user's gaze lingering on a restaurant card for more than a preset time (e.g., 2 seconds) is considered a browsing action); clicking restaurant content (including clicking the "View Details" button, clicking card titles or image areas, triggering page redirects or pop-ups); adding restaurant content to favorites (clicking the "Add to Favorites" button, recording both successful and unsuccessful additions); and submitting reviews or ratings on the recommended content details page or through a dedicated review portal. The module records the type, time, corresponding restaurant content identifier, and user device information (such as browser type, screen resolution, and other non-privacy sensitive information) of these behaviors in real time.

[0144] Step S155: Record the time of each operation, the corresponding catering content identifier, and the operation type to form an initial interaction record.

[0145] The user interaction behavior capture module records each monitored operation behavior in the form of a structured log, forming an initial interaction record. Each initial interaction record contains the following fields: the time of the operation behavior, accurate to milliseconds, using UTC time format; the corresponding food and beverage content identifier, i.e., the unique code of the food and beverage content targeted by the operation behavior within the platform; the operation behavior type, represented by a predefined enumeration value, such as "VIEW" for "browse", "CLICK" for "click", "COLLECT" for "favorite", and "RATE" for "rate"; operation behavior parameters, recording additional information according to the behavior type, such as the dwell time of the browsing behavior, the specific location coordinates of the clicking behavior, the rating of the rating behavior, and the text content (the text content must be anonymized to remove sensitive personal information). The initial interaction records are temporarily stored in the user's local cache. When a certain number (e.g., 10 records) are reached or after a certain interval (e.g., 30 seconds), they are uploaded in batches to the platform's interaction behavior data receiving server.

[0146] Step S156: Extract intent-related information corresponding to the operation behavior from the initial interaction record. The intent-related information includes the user's preference for category selection, price focus, and service demand for catering content, forming new catering interaction intent data.

[0147] After receiving the initial interaction records, the platform's interactive behavior data receiving server extracts intent-related information from each record. For category selection preference, the system associates restaurant content with its category (e.g., Japanese cuisine, Chinese stir-fry), and counts the frequency of user actions on different categories within a unit of time; higher frequency indicates a stronger category selection preference. For price focus preference, the system obtains price information from the restaurant content details and combines this with user clicks and favorites to determine the user's level of interest in different price ranges. For example, multiple clicks on mid-to-high-priced restaurant content indicate a mid-to-high price focus preference. For service demand preference, the system analyzes whether user actions involve service-related information, such as clicking on "delivery policy" links or mentioning keywords like "good environment" in reviews, extracting the user's demand preference for delivery services, dining environment, and promotional activities. The extracted category selection preference, price focus preference, and service demand preference, along with the corresponding operation time and behavior type, are integrated into structured new restaurant interaction intent data. The data format remains consistent with the original restaurant interaction intent data for subsequent integration.

[0148] Step S157: Format the new catering interaction intent data, add the formatted new catering interaction intent data to the original catering interaction intent data, and update the intent time series and intent node association information of the original catering interaction intent data.

[0149] The new restaurant interaction intent data undergoes format processing, including data cleaning (removing duplicate records and correcting abnormal timestamps), field standardization (unifying category names and price range division standards), and data compression (using an efficient data encoding format for storage). After processing, the new restaurant interaction intent data is added to the user's existing restaurant interaction intent data set through database append or update operations. The original restaurant interaction intent data contains an intention time series that is a list of interaction behavior records arranged chronologically. After adding new data, the new records need to be inserted into the corresponding positions of the time series according to their timestamps to maintain the order of the time series. Simultaneously, the intent node association information is updated, recording which intent node each new interaction behavior record belongs to (usually associated with the current intent node, or with the new intent node if the intent changes), and updating the statistical information of the interaction behavior list corresponding to the intent node (such as interaction behavior frequency, frequency of intent tendency occurrence, etc.).

[0150] Step S158: Extract the latest intent change nodes and intent duration information from the updated catering interaction intent data, adjust the intent evolution link in the user catering intent evolution model, and update the core intent tendency and performance characteristics of the intent nodes.

[0151] Step S1581: Extract the latest interaction behavior records from the updated catering interaction intent data and arrange them in chronological order to form the latest intent time series.

[0152] From the updated restaurant interaction intent data, we filter out interaction behavior records from the most recent period (e.g., the past 24 hours or since the last model update). These records include user actions on recommended restaurant content and other restaurant-related content. These records are then arranged in ascending order of the interaction timestamp to form the latest intent time series. This latest intent time series is a continuation and supplement to the original intent time series, reflecting the latest user interaction dynamics.

[0153] Step S1582: Analyze the changes in intent tendencies in the latest intent time series. If an intent tendency that is different from the core intent tendency of the current intent node appears, and the frequency of the interaction behavior corresponding to the intent tendency that is different from the core intent tendency of the current intent node reaches a preset frequency threshold, mark the corresponding time point as a new intent change node.

[0154] Analyze the changes in user intent tendencies in the latest intent time series by comparing the intent tendency corresponding to each interaction record with the core intent tendency of the current intent node (e.g., the current core intent tendency is a preference for Japanese cuisine). If it is found that the user begins to exhibit intent tendencies different from the current core intent tendency, such as repeatedly browsing Korean food content such as Korean BBQ and bibimbap, and the frequency of these Korean food-related interactions reaches a preset frequency threshold (this threshold is set based on the average frequency of the user's historical interactions and the sensitivity to intent changes, for example, 50% of the daily average frequency of interactions related to the core intent tendency of the current intent node), then it is determined that the user's intent tendency has changed significantly. Find the time point in the latest intent time series where this different intent tendency first appears in a concentrated manner, and mark this time point as a new intent change node, serving as a new starting point for intent evolution.

[0155] Step S1583: Starting from the new intent change node, determine the new intent duration period, and collect the interactive behavior records and corresponding intent tendencies within the new intent duration period.

[0156] A new intent duration period is determined starting from the newly marked intent change node. The end time of the intent duration period is tentatively set to the current time; if a new intent change node appears later, the end time will be the time of the new node. Within the new intent duration period, all interaction records are collected, including user browsing, clicking, saving, and rating of Korean cuisine and other food categories. Simultaneously, corresponding intent tendencies are extracted for each record, such as preference for Korean cuisine, acceptance of mid-to-high price ranges, and demand for delivery services, and the time and frequency of each intent tendency are recorded.

[0157] Step S1584: Calculate the frequency of occurrence of each intent tendency within the new intent duration period, and select the intent tendency with the highest frequency of occurrence as the core intent tendency of the new intent node.

[0158] The frequency of occurrence of all intent tendencies recorded within the new intent duration period is calculated. The frequency calculation method is to sum the number of interaction behaviors for the same type of intent tendency (such as a preference for Korean cuisine). For example, within the new period, the interaction behaviors corresponding to the Korean cuisine preference are 8 views, 6 clicks, and 2 favorites, for a total frequency of 16; the interaction behaviors corresponding to the Chinese stir-fry preference are 3 views and 1 click, for a total frequency of 4; other categories have lower frequencies. The intent tendency with the highest occurrence frequency (Korean cuisine preference) is selected as the core intent tendency of the new intent node.

[0159] Step S1585: Statistically analyze the frequency of interactive behaviors corresponding to the core intent tendency of the new intent node, the degree of correlation with the interactive behaviors, and the degree of potential impact on subsequent intent changes to form the performance characteristics of the new intent node.

[0160] The frequency of interactive behaviors corresponding to the core intent tendency (Korean cuisine category preference) of the new intent node is statistically analyzed, i.e., the total number of interactive behaviors (e.g., 16 times) within the duration of the new intent. The degree of correlation with interactive behaviors is measured by calculating the proportion of deep interactive behaviors (e.g., clicks, favorites, reviews) in the total interactive behaviors corresponding to the core intent tendency. The higher the proportion, the stronger the correlation. For example, out of 16 interactive behaviors, clicks and favorites account for 8 times, a proportion of 50%, indicating a moderate degree of correlation. The potential impact on subsequent intent changes is assessed by analyzing the stability and diversity of the core intent tendency. Stability is calculated by calculating the standard deviation of the intent tendency frequency (the smaller the fluctuation, the more stable). Diversity is determined by counting the number of subcategories under this tendency (e.g., Korean BBQ, bibimbap, army stew, etc.). The more subcategories there are, the higher the diversity and the greater the potential impact. A potential impact score (e.g., high, medium, low) is given by combining stability and diversity. The above statistical results are integrated to form a description of the performance characteristics of the new intent node.

[0161] Step S1586: Add the new intent node to the end of the intent evolution link of the user's catering intent evolution model, calculate the intent association strength between the new intent node and the previous intent node, and use the intent association strength as a connection edge to connect the new intent node and the previous intent node.

[0162] The newly constructed intent node (Korean cuisine category preference intent node) is added to the end of the intent evolution chain of the user's dining intent evolution model, making it the newest node in the chain. Then, the intent association strength between the new intent node and the previous intent node (Japanese cuisine category preference intent node) is calculated. The calculation method is the same as in step S124, that is, the degree of overlap of intent tendencies (such as whether the price acceptance tendency is the same), the smoothness of intent change (node ​​interval duration), and the continuity of interaction behavior (such as whether some Japanese cuisine interaction behavior continues) are statistically analyzed to obtain the intent association strength level (such as medium). The new intent node is connected to the previous intent node with the connection edge marked with the association strength level, and the structure of the intent evolution chain is updated.

[0163] Step S1587: If the latest intent time series does not show any intent trends that match the new intent change nodes, update the interaction behavior list of the current intent node and add the latest interaction behavior record; recalculate the frequency of occurrence of each intent trend in the interaction behavior list of the current intent node. If the frequency ratio of the core intent trend of the current intent node drops below the preset ratio threshold, redetermine the core intent trend of the current intent node; recalculate the performance characteristic parameters of the current intent node and update the performance characteristic description of the current intent node; adjust the intent association strength between the current intent node and the previous intent node so that the intent association strength between the current intent node and the previous intent node reflects the latest interaction behavior continuation.

[0164] If the user's intent tendency in the latest intent time series has not changed significantly, i.e., no node matching the new intent change appears, then the interaction behavior list of the current intent node is updated, and the latest interaction behavior record is added to the list. The frequency of each intent tendency in the current intent node's interaction behavior list is recalculated, and the proportion of the core intent tendency's frequency to the total frequency is calculated. If this proportion drops below a preset threshold (e.g., 50%), it indicates that the core intent tendency may have shifted, and the core intent tendency needs to be redefined, selecting the intent tendency with the highest current frequency proportion as the new core intent tendency. At the same time, the performance characteristic parameters of the current intent node are recalculated, such as intent duration (adding the time span of the latest interaction behavior), interaction behavior frequency (adding the number of the latest record), and intent stability index (recalculated based on the latest frequency fluctuations), and the performance characteristic description is updated. In addition, based on the latest interaction behavior continuity (e.g., whether the interaction behavior of the current intent node is still continuous with the interaction behavior of the previous node), the intent association strength between the current intent node and the previous intent node is recalculated, and the strength level of the connection edge is adjusted to ensure that the intent evolution link accurately reflects the latest state of the user's intent.

[0165] Step S159: Call the update module of the intent supply time sequence adaptation logic, input the new catering interaction intent data and the corresponding supply time sequence data into the update module of the intent supply time sequence adaptation logic, adjust the weight parameters and supply cycle stage matching tag allocation conditions in the adaptation rules; record the update content of the user catering intent evolution model and intent supply time sequence adaptation logic to form an update log, which includes the update time, the adjusted parameter content and the update basis.

[0166] The update module, which handles the timing adaptation logic for intent-based supply, receives new data and adjusts the adaptation logic parameters. New food and beverage interaction intent data (including the user's latest intent tendencies and interactive behaviors) and corresponding food and beverage content supply timing data (such as the latest supply cycle stage characteristics and supply volume trends) are input into the update module. Based on the new data, the update module adjusts the weight parameters (such as the weight coefficient of category preference ratio) and the supply cycle stage matching tag allocation conditions (such as the adaptation threshold for the main adaptation supply stage) in the adaptation rules using a preset parameter adjustment algorithm (such as a gradient descent-based parameter optimization algorithm). During the adjustment process, the goal is to improve the fit between intent-based supply adaptation and actual user interaction behavior, finding the optimal parameter combination through iterative calculation.

[0167] After the update is completed, the update content of the user's dining intention evolution model and the timing adaptation logic of intention supply is recorded, forming an update log. The update log includes: the update time, accurate to the second; the adjusted parameters, such as "adjusting the category preference weight coefficient from 0.2 to 0.25" and "adjusting the adaptation threshold of the main adaptation supply stage from 0.7 to 0.75"; and the basis for the update, such as "increasing the category preference weight based on a 30% increase in the frequency of user interaction with Korean cuisine in the past 7 days" and "accurately verified by historical data, increasing the threshold improves the adaptation success rate by 5%". The update log is stored in the platform's version control system and is used to trace the history of parameter changes, troubleshoot problems, and perform model optimization analysis.

[0168] Figure 2 This application illustrates a restaurant recommendation system 100 for a digital multimedia online platform, including a processor 1001, a memory 1003, and program code stored in the memory 1003. The processor 1001 executes the program code to implement the steps of a restaurant recommendation method for a digital multimedia online platform. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the restaurant recommendation system 100 may further include a transceiver 1004, which can be used for data interaction between this restaurant recommendation system and other restaurant recommendation systems applied to the digital multimedia online platform, such as sending and / or receiving data. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of this restaurant recommendation system 100 for a digital multimedia online platform does not constitute a limitation on the embodiments of this application.

[0169] The memory 1003 is used to store program code for executing the embodiments of this application, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the foregoing method embodiments.

[0170] This application provides a computer-readable storage medium storing program code, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0171] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. A method for restaurant recommendation applied to a digital multimedia online platform, characterized in that, The method includes: The system acquires user interaction intent data on a digital multimedia online platform and supply time-series data of catering content within the platform. The interaction intent data includes the intent tendency, intent change nodes, and intent duration corresponding to user interaction behavior. The supply time-series data of catering content includes the supply cycle stages, supply volume change trends, and supply matching intent tags. The intent change nodes are the time points in the intent time series where the intent tendency changes. The supply cycle stages include the new product launch stage, hot-selling stage, stable supply stage, and delisting preparation stage of catering content. The supply matching intent tags are pre-set according to the characteristics of catering content and are used to match different user intent tendencies. Based on the intent change patterns in the catering interaction intent data and the supply time series data of catering content, a user catering intent evolution model is constructed. The user catering intent evolution model takes the intent nodes of users at different time periods as the core and forms an intent evolution link by connecting them through intent association strength. The intent association strength is obtained by comprehensively considering the degree of overlap of intent tendencies between two adjacent intent nodes, the smoothness of intent change, and the continuity of interaction behavior. By combining the user's dining intention evolution model with the supply time series data of dining content, an intention supply time series adaptation logic is established. The intention supply time series adaptation logic is used to associate the user's intention evolution stage with the supply cycle stage of dining content. Based on the intention-supply time sequence adaptation logic, select catering content from the platform's catering content library that is in the matching supply cycle stage and fits the user's current intention evolution stage, and generate a dynamic catering recommendation scheme. The dynamic restaurant recommendation scheme is pushed to the user display interface of the digital multimedia online platform. At the same time, the new restaurant interaction intent data of the user on the dynamic restaurant recommendation scheme is captured and integrated into the original restaurant interaction intent data to update the user's restaurant intent evolution model and intent supply time sequence adaptation logic. The step of combining the user's dining intention evolution model with the time-series data of dining content supply to establish an intention supply time-series adaptation logic includes: Analyze the supply time series data of catering content and extract the supply cycle stage of each catering content; The supply characteristics of catering content in each supply cycle stage are statistically analyzed. The supply characteristics include the supply volume change curve of catering content, the user access volume change trend, and the platform recommendation resource allocation ratio in that supply cycle stage. The core intent tendency and performance characteristics of each intent node are extracted from the user's dining intent evolution model to form an intent feature set, which includes the periodic identifier of the intent node, core intent tendency parameters and performance characteristic parameters. The intent feature set of each intent node is correlated with the supply cycle stage features of each catering content to generate intent supply fit. The correlation analysis is specifically achieved by comparing whether the core intent tendency in the intent features overlaps with the supply matching intent tags, assessing whether the user's intent needs match the supply of catering content, and judging whether the intent duration of the intent node is synchronized with the supply cycle stage of the catering content in time. Based on the supply adaptation degree of the intent, a corresponding supply cycle stage matching label is assigned to each intent node. The supply cycle stage matching label includes the primary supply adaptation stage, the secondary supply adaptation stage, and the adaptation priority. The matching frequency of all intent nodes with each supply cycle stage is counted to generate an intent supply matching frequency distribution table. The intent supply matching frequency distribution table records the total number of times each supply cycle stage is matched, the type distribution of the corresponding intent nodes, and the average fit. Based on the frequency distribution table of intent supply matching, the adaptation rules for each intent node and the corresponding supply cycle stage are determined. The adaptation rules include the adaptation degree calculation standard, the supply cycle stage matching label allocation conditions, and the adaptation priority ranking basis. Collect historical user intent evolution data and corresponding supply time series data, use historical user intent evolution data and corresponding supply time series data to verify the effectiveness of adaptation rules, and adjust adaptation rule parameters by comparing the degree of fit between the adaptation results output by the adaptation rules and the actual historical interaction effects. The verification and adjustment of the adaptation rules, the calculation method of the intention supply adaptation degree and the supply cycle stage matching label allocation mechanism are integrated to form the basic framework of the intention supply time sequence adaptation logic. An adaptation logic update module is added to the basic framework of intent supply timing adaptation logic. This module is used to receive new interactive intent data and supply timing data, dynamically adjust the adaptation rules and adaptation degree calculation parameters, and complete the establishment of intent supply timing adaptation logic.

2. The restaurant recommendation method applied to a digital multimedia online platform according to claim 1, characterized in that, The aforementioned model for the evolution of user dining intentions, based on the patterns of intention changes in dining interaction intention data and the time-series data of dining content supply, includes: Analyze the restaurant interaction intent data and extract the intent tendency corresponding to each interaction behavior record. The intent tendency includes the user's preference tendency for restaurant categories, acceptance tendency for restaurant prices, and demand tendency for restaurant services. All interaction records are arranged in chronological order to form an intent time series. The time points in the intent time series where the intent tendency changes are marked as intent change nodes. The time interval between two adjacent intention change nodes is taken as an intention duration period. The intention time series is divided into multiple consecutive intention duration periods, and each intention duration period corresponds to a stable intention tendency. The performance characteristics of intent tendencies within each intent duration period are statistically analyzed. These performance characteristics include the frequency of occurrence of intent tendencies within the intent duration period, the degree of correlation with interaction behavior, and the degree of influence on subsequent intent changes. Based on the intent tendency and performance characteristics of each intent duration, intent nodes are constructed. Each intent node includes a period identifier, core intent tendency, performance characteristic description, and a corresponding list of interactive behaviors. Calculate the intent association strength between two adjacent intent nodes; Using intent association strength as the connection edge, all intent nodes are connected in chronological order of intent duration to form the initial intent evolution link; Extract the supply cycle stage features corresponding to each intent node from the supply time series data of catering content, and add the supply cycle stage features as supplementary information to the corresponding intent node to enrich the feature dimensions of the intent node. Traverse all connection edges in the initial intent evolution link, and adjust the intent association strength according to the supplementary supply cycle stage characteristics so that the intent association strength reflects both the intent change pattern and the supply time sequence matching degree. Remove the connection edges and corresponding intent nodes in the initial intent evolution link that are lower than the set threshold and have no subsequent interaction data support, and retain the intent nodes and connection edges that meet the intent association strength requirements or have subsequent interaction data support to form an intent evolution link. Based on the intent evolution chain, a model update interface, an intent node query interface, and an association strength calculation interface are added to construct a user catering intent evolution model.

3. The restaurant recommendation method applied to a digital multimedia online platform according to claim 1, characterized in that, The step of selecting catering content from the platform's catering content library that is in the matching supply cycle stage and matches the user's current intention evolution stage based on the intent supply time sequence adaptation logic, and generating a dynamic catering recommendation scheme, includes: Extract the latest intent node from the user's dining intent evolution model as the current intent node, and determine the core intent tendency and main supply stage corresponding to the current intent node. Invoke the intent supply timing adaptation logic, query the supply cycle stage characteristics that match the core intent tendency of the current intent node, and obtain the target supply cycle stage characteristics; Extract all catering content that is in the target supply cycle stage from the platform's catering content library to form an initial catering content set; Extract the supply characteristics and supply matching intent tags of each food and beverage item in the initial food and beverage content set, and compare the supply matching intent tags with the core intent tendency of the current intent node; Retain catering content whose supply matching intent tags overlap with the core intent tendency of the current intent node, forming an intermediate catering content set; Extract a list of restaurant content that the user has interacted with from the restaurant interaction intent data, remove restaurant content from the intermediate set that exceeds a preset duplication threshold in terms of core intent matching with the restaurant content that the user has interacted with, and obtain the final set of restaurant content. Invoke the intent supply timing adaptation logic to calculate the intent supply adaptation degree between each piece of catering content in the final catering content set and the current intent node; The food and beverage items in the final food and beverage content set are sorted in descending order of their intended supply suitability to form an ordered food and beverage content list. Select the first few items from the ordered catering content list, and extract the display information for each item. The display information includes catering content images, catering details descriptions, supply cycle stage identifiers, and intent matching instructions. The selected catering content display information is integrated according to the sorting order of the catering content list, and the recommendation generation time, current intent node identifier and target supply cycle stage identifier are added to form a dynamic catering recommendation scheme. Add update trigger conditions to the dynamic recommendation scheme for catering. The update trigger conditions include new user interaction behavior, changes in the current intent node, and updates in the supply cycle stage.

4. The restaurant recommendation method applied to a digital multimedia online platform according to claim 3, characterized in that, The intent supply timing adaptation logic calculates the intent supply adaptation degree between each piece of catering content in the final catering content set and the current intent node, including: Extract individual food items from the final set of food content, obtain the supply time series data corresponding to the food content, and parse the supply cycle stage characteristic parameters. The supply cycle stage characteristic parameters include the peak time of supply volume, the growth rate of user visits, and the proportion of platform recommended resources. Convert the characteristic parameters of the supply cycle stage into a standardized supply characteristic vector; Extract core intent tendency parameters and performance characteristic parameters from the current intent node. The core intent tendency parameters include category preference ratio, price acceptance range ratio, and service demand ratio. The performance characteristic parameters include intent duration, interaction frequency, and intent stability index. Convert the core intent tendency parameters and performance characteristic parameters into standardized intent feature vectors; Call the adaptation degree calculation module in the intent supply timing adaptation logic, and obtain the pre-stored feature weight allocation table in the adaptation calculation module. The feature weight allocation table contains the weight coefficients corresponding to different feature dimensions, and the weight coefficients are dynamically adjusted according to the historical adaptation effect. For each pair of corresponding dimension features in the standardized supply feature vector and the standardized intent feature vector, calculate its matching score; according to the weight coefficient of the corresponding dimension in the feature weight allocation table, calculate the weighted average of the matching scores of all dimensions to obtain the intent supply fit degree between the catering content and the current intent node. Repeat the above steps to calculate the intent supply fit of all catering content in the final catering content set with the current intent node, associate and store the intent supply fit of all catering content with the corresponding catering content identifier, and form a fit statistics list.

5. The restaurant recommendation method applied to a digital multimedia online platform according to claim 1, characterized in that, The process of pushing the dynamic restaurant recommendation scheme to the user's display interface on the digital multimedia online platform, while simultaneously capturing new restaurant interaction intent data of the user regarding the dynamic restaurant recommendation scheme, and integrating the new restaurant interaction intent data into the original restaurant interaction intent data, is used to update the user's restaurant intent evolution model and intent supply time sequence adaptation logic, including: Call the data transmission interface of the digital multimedia online platform to transmit the catering content display information, supply cycle stage identifiers and intent matching instructions in the catering dynamic recommendation scheme to the data transmission interface of the display interface. After receiving data, the display interface data transmission interface organizes the data according to the preset recommended content layout rules of the digital multimedia online platform to generate display interface rendering data. The recommended content layout rules include the display order of catering content, information layout format and interactive entry settings. The digital multimedia online platform display interface sends the rendering data of the display interface to the user's terminal, and displays the catering content in the recommended content area of ​​the digital multimedia online platform display interface according to the sorting order of the catering dynamic recommendation scheme. The user interaction behavior capture module of the digital multimedia online platform is activated to monitor the user's operation behavior in the recommended content area in real time. The operation behavior includes browsing food and beverage content, clicking on food and beverage content, collecting food and beverage content, and rating food and beverage content. Record the time of each operation, the corresponding food and beverage content identifier, and the type of operation to form an initial interaction record; Extract intent-related information corresponding to the operation behavior from the initial interaction record. The intent-related information includes the user's preference for the category of catering content, price focus, and service demand, forming new catering interaction intent data. The new catering interaction intent data is formatted and added to the original catering interaction intent data, and the intent time series and intent node association information of the original catering interaction intent data are updated. Extract the latest intent change nodes and intent duration information from the updated catering interaction intent data, adjust the intent evolution link in the user catering intent evolution model, and update the core intent tendency and performance characteristics of intent nodes. Call the update module of the intent supply time sequence adaptation logic, input the new catering interaction intent data and the corresponding supply time sequence data into the update module of the intent supply time sequence adaptation logic, and adjust the weight parameters and supply cycle stage matching tag allocation conditions in the adaptation rules. The update log records the user's dining intention evolution model and the timing adaptation logic of intention supply, and includes the update time, the adjusted parameters, and the basis for the update.

6. The restaurant recommendation method applied to a digital multimedia online platform according to claim 2, characterized in that, The calculation of the intent association strength between two adjacent intent nodes includes: Extract two adjacent intent nodes and denote them as the preceding intent node and the following intent node, respectively. Extract the core intent tendency from the preceding intent nodes to form a set of preceding intent tendencies; extract the interaction behavior type and interaction frequency from the interaction behavior list corresponding to the preceding intent nodes to form a set of preceding interaction behaviors. Extract the core intent tendency from the subsequent intent nodes to form a set of subsequent intent tendencies; extract the interaction behavior type and interaction frequency from the interaction behavior list corresponding to the subsequent intent nodes to form a set of subsequent interaction behaviors. The number of intents with the same type and identical content in the preceding intent set and the subsequent intent set is recorded as the number of overlapping intents. The degree of overlap of intents is obtained by calculating the proportion of the number of overlapping intents to the total number of preceding intent sets. Calculate the time interval between the end time of the intent duration period of the preceding intent node and the start time of the intent duration period of the subsequent intent node, and denot it as the node interval duration; determine the intent change smoothing coefficient based on the node interval duration, the shorter the node interval duration, the larger the intent change smoothing coefficient. The number of interaction behaviors of the same type in the preceding and subsequent interaction behavior sets is recorded as the number of continuation interactions; the proportion of the number of continuation interactions to the total number of preceding interaction behaviors is calculated to obtain the interaction behavior continuation degree. Threshold standards are set for evaluating intent overlap, intent change smoothness coefficient, and interaction behavior continuity; intent overlap, intent change smoothness coefficient, and interaction behavior continuity are compared with their respective threshold standards to obtain their respective sub-evaluation results; based on the sub-evaluation results, the intent association strength level between two adjacent intent nodes is comprehensively judged.

7. The restaurant recommendation method applied to a digital multimedia online platform according to claim 1, characterized in that, The process of collecting historical user intent evolution data and corresponding supply time-series data, verifying the effectiveness of adaptation rules using this data, and adjusting adaptation rule parameters by comparing the adaptation results output by the adaptation rules with the actual historical interaction effects includes: The historical data storage module of the digital multimedia online platform extracts the catering interaction intent data of multiple historical users to form a historical intent evolution dataset. The historical intent evolution dataset includes the intent node sequence, intent association strength and intent change pattern of each historical user. Extract the time-series data of catering content supply corresponding to the historical intent evolution dataset to form a historical supply time-series dataset. The historical supply time-series dataset includes the supply cycle stage, supply characteristics and supply matching intent tags of catering content in each historical time period. Input a single historical intent evolution data point from the historical intent evolution dataset into the basic framework of intent supply time-series adaptation logic, and output the corresponding supply cycle stage matching result and intent supply adaptation degree according to the adaptation rules to form the prediction adaptation result. Extract the actual user interaction records corresponding to the historical intent evolution data from the platform's historical data. The actual user interaction records include the food and beverage content actually selected by the user, the interaction time, and the interaction frequency. The number of successful matches between the supply cycle stage corresponding to the actual catering content selected by the user in the statistical prediction and adaptation results is recorded as the number of successful matches. The total number of all recommended supply cycle stages in the statistical prediction and adaptation results is recorded as the total number of predictions. The proportion of successful matches to the total number of predictions is calculated to obtain the adaptation success rate of a single historical intent evolution data. Repeat the above steps to calculate the adaptation success rate of all historical intent evolution data in the historical intent evolution dataset, and take the average of the adaptation success rates of all historical intent evolution data to obtain the overall adaptation success rate. If the overall adaptation success rate reaches the adaptation success rate threshold, the current parameters of the adaptation rule are deemed valid; if the overall adaptation success rate does not reach the adaptation success rate threshold, the differences between the predicted adaptation results and the actual user interaction records are analyzed. The differences include the intention supply adaptation degree calculation deviation, the supply cycle stage matching tag allocation deviation, and the adaptation priority ranking deviation. Based on the differences, adjust the adaptation degree calculation standard parameters, supply cycle stage matching label allocation condition parameters, and adaptation priority sorting parameters in the adaptation rules. After adjustment, re-enter historical intent evolution data and corresponding supply time series data for verification until the overall adaptation success rate reaches the adaptation success rate threshold, thus completing the adjustment of the adaptation rule parameters.

8. The restaurant recommendation method applied to a digital multimedia online platform according to claim 5, characterized in that, The process involves extracting the latest intent change nodes and intent duration information from the updated dining interaction intent data, adjusting the intent evolution chain in the user dining intent evolution model, and updating the core intent tendencies and performance characteristics of intent nodes, including: Extract the latest interaction behavior records from the updated catering interaction intent data and arrange them in chronological order to form the latest intent time series; Analyze the changes in intent tendencies in the latest intent time series. If an intent tendency that is different from the core intent tendency of the current intent node appears, and the frequency of the interaction behavior corresponding to the intent tendency that is different from the core intent tendency of the current intent node reaches a preset frequency threshold, mark the corresponding time point as a new intent change node. Starting from the new intent change node, determine the new intent duration period, and collect the interactive behavior records and corresponding intent tendencies within the new intent duration period. Calculate the frequency of occurrence of each intent tendency within the duration of the new intent, and select the intent tendency with the highest frequency of occurrence as the core intent tendency of the new intent node; The frequency of interactive behaviors corresponding to the core intent tendency of new intent nodes, the degree of correlation with interactive behaviors, and the degree of potential impact on subsequent intent changes are statistically analyzed to form the performance characteristics of new intent nodes. Add the new intent node to the end of the intent evolution chain of the user's dining intent evolution model, calculate the intent association strength between the new intent node and the previous intent node, and use the intent association strength as a connection edge to connect the new intent node and the previous intent node. If the latest intent time series does not show any nodes that match the new intent change, update the interaction behavior list of the current intent node and add the latest interaction behavior record. Recalculate the frequency of occurrence of each intent tendency in the current intent node's interaction behavior list. If the frequency of occurrence of the core intent tendency of the current intent node drops below the preset threshold, redetermine the core intent tendency of the current intent node. Recalculate the performance feature parameters of the current intent node and update the performance feature description of the current intent node; Adjust the intent association strength between the current intent node and the previous intent node so that the intent association strength between the current intent node and the previous intent node reflects the latest interaction behavior continuation.

9. A restaurant recommendation system applied to a digital multimedia online platform, characterized in that, The invention includes a processor and a computer-readable storage medium storing machine-executable instructions that, when executed by the processor, implement the restaurant recommendation method for a digital multimedia online platform as described in any one of claims 1-8.

Images