An individualized menu pushing system based on user portrait
By combining multi-dimensional, layered user profiles with real-time scenario data, the system addresses the issues of insufficient accuracy and personalization in restaurant recommendation systems, enabling precise and personalized menu pushes to meet diverse user dining needs.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SUZHOU CENTENNIAL VOCATIONAL COLLEGE
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-19
AI Technical Summary
Existing restaurant recommendation systems suffer from low accuracy, insufficient personalization, and poor real-time performance, failing to meet the diverse dining needs of users and the refined operational requirements of businesses.
It adopts a multi-dimensional, layered user profile construction, including basic attributes, eating habits, scenario needs, and consumption behavior tags. Combined with real-time scenario data, it dynamically adjusts dining intentions and achieves precise menu push through dish selection, intention mining, and menu recommendation units.
It improves the accuracy and efficiency of menu recommendations, better adapts to users' core needs and real-time scenarios, and enhances the personalization and real-time nature of menu recommendations.
Smart Images
Figure CN122240926A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent recommendation technology for catering, and in particular to a personalized menu push system based on user profiles. Background Technology
[0002] With the digital and intelligent development of the catering industry, intelligent menu recommendation systems have become an important means for catering businesses to improve user experience and increase operational efficiency. Personalized menu push is an intelligent system that relies on big data and intelligent algorithms to first build exclusive user profiles for catering users, and then combine profile characteristics, consumption scenarios and other factors to accurately push menus / dishes that match users' dietary preferences and needs. It is mainly used in scenarios such as food delivery platforms, private domains of catering brands, and smart restaurants. Its core is to achieve "personalized menu recommendations" for each user, which is different from the traditional uniform menu display.
[0003] Currently, most restaurant recommendation systems on the market match dishes based on single user tags, resulting in problems such as low recommendation accuracy, insufficient personalization, and poor real-time performance. They can no longer meet the diverse dining needs of users and the refined operational requirements of businesses. The main shortcomings of existing technologies are as follows: 1. User profiles are incomplete and lack dimensionality, failing to be layered according to dining scenarios. Tags lack synergy, failing to accurately reflect users' dietary preferences, scenario needs, and consumption habits. 2. Using a full-dish pool for filtering is inefficient. Furthermore, the filtering of dining intentions is directly compared with tags, failing to uncover deeper user dining intentions and failing to dynamically adjust the intention judgment results based on real-time scenario data, leading to a disconnect between the recommendation direction and the user's current dining needs. Summary of the Invention
[0004] The purpose of this invention is to provide a personalized menu push system based on user profiles to solve the problem of poor adaptability of recommended dishes in the above-mentioned restaurant recommendation system. It has the advantages of multi-dimensional and layered user profiles, accurate mining of user dining intentions, and more accurate menu recommendations.
[0005] Firstly, to achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0006] A personalized menu push system based on user profiles includes:
[0007] The data acquisition unit is used to collect multi-dimensional data from users, which is obtained by parsing the ordering system, self-filling page, social platform interface and group buying platform interface.
[0008] The user profiling unit constructs user profile tags based on the multi-dimensional data. The user profile tags are four-layer tags, which include basic attribute tags, dietary habit tags, scenario demand tags, and consumption behavior tags. The dietary habit tags include multiple reference factors.
[0009] The dish selection unit includes a scenario determination module, used to set an initial recommendation range based on the basic attribute tags and the scenario requirement tags; a filtering module, used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; and a basic matching module, used to select matching dishes from the remaining dishes in the initial recommendation range based on the four layers of tags and add them to the basic dish pool.
[0010] The intent mining unit is used to acquire real-time scene data, which includes time period, weather and user current status, and to generate dining intent based on the four-layer tags and the real-time scene data.
[0011] The menu recommendation unit is used to select dishes from the basic menu pool based on the user's dining intention, combine them to generate a menu, and push it to the user.
[0012] Preferably, each tag in the user profile unit includes multiple reference factors, specifically:
[0013] The reference factors for basic attribute tags include, in order, commonly used regions, age ranges, and gender information;
[0014] The reference factors for dietary habit labels include, in order of importance, taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences.
[0015] The reference factors for scenario demand tags include, in order, dining scenario, number of diners, real-time dietary needs, and dining time.
[0016] The reference factors for consumer behavior tags include, in order of importance, the average order value range, the frequency of ordering, the dishes that are repeatedly purchased, and the preference for dish pairings.
[0017] Preferably, the method for adding matching dishes to the basic dish pool by filtering them in the basic matching module includes:
[0018] Calculate the matching scores of the dishes with each tag: basic attribute tag score S1, dietary habit tag score S2, scenario demand tag score S3, and consumption behavior tag score S4.
[0019] The weights for each label are preset as a, b, c, and d, respectively.
[0020] The overall matching score of the dishes is calculated using the following formula: S = a × S1 + b × S2 + c × S3 + d × S4;
[0021] A preset filtering threshold M is set. If the M ≤ S for a dish, then the dish is added to the basic dish pool.
[0022] Preferably, the intent mining unit includes a candidate intent module, used to obtain candidate intents, and the method for obtaining candidate intents is as follows:
[0023] Establish a dining intent library, which includes multiple intent units, and the dining intents output by the intent mining unit can be composed of one or more intent units;
[0024] The dietary habit tag and scenario requirement tag in the four-layer label are associated with each intent unit, and the association matching score is calculated.
[0025] A preset candidate filtering threshold is set, and intent units with a correlation matching score greater than the candidate filtering threshold are set as candidate intents.
[0026] Preferably, the intent mining unit further includes an auxiliary intent module, which is used to calculate the matching score between the candidate intent and the basic attribute tag and consumption behavior tag in the four-layer tag, and use it as an auxiliary matching score. The association matching score and the auxiliary matching score are weighted and fused to obtain a comprehensive score, and the candidate intent with the highest comprehensive score is selected as the core intent.
[0027] Preferably, the intent mining unit includes an intent fusion generation module, which is used to calculate the matching degree by weighted fusion of core intent and real-time scene data, and obtain the core intent with the highest matching degree as the final dining intent.
[0028] Preferably, the menu recommendation unit includes a dish mapping module, used to convert dining intentions into quantifiable dish filtering tags and establish a mapping table between dining intentions and dish tags; a scene feature module, used to establish two types of scenes: single-person set meals and multi-person set meals, with each scene including the structure and quantity of dishes; and a dish dynamic combination module, used to select dishes with high matching degree with dining intentions from the basic dish pool according to the mapping table and generate combined set meals according to the scene feature module.
[0029] Preferably, the system also includes a personalized verification unit, which is used to set an update time to periodically verify the four-layer tags in the user profile unit.
[0030] Secondly, to achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0031] A personalized menu push method based on user profiles includes the following steps:
[0032] Step S1, data collection, is used to collect multi-dimensional data of users. The multi-dimensional data is obtained through parsing the ordering system, the self-filling page, the social platform interface, and the group buying platform interface.
[0033] Step S2: Construct a user profile. Based on the multi-dimensional data, construct user profile tags. The user profile tags are four-layer tags, which include basic attribute tags, dietary habit tags, scenario demand tags, and consumption behavior tags in sequence. The dietary habit tags include multiple reference factors.
[0034] Step S3, dish selection, includes a scenario determination module, used to set an initial recommendation range based on the basic attribute tags and the scenario requirement tags; a filtering module, used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; and a basic matching module, used to select matching dishes from the remaining dishes in the initial recommendation range based on the four layers of tags and add them to the basic dish pool.
[0035] Step S4, Intent mining, obtaining real-time scene data, the real-time scene data including time period, weather and user current status, and generating dining intent based on the four-layer tags and the real-time scene data;
[0036] Step S5, Menu Recommendation: Based on the dining intention, select dishes from the basic menu pool, combine them to generate a menu, and push it to the user.
[0037] Preferably, each label in step S2 includes multiple reference factors, specifically:
[0038] The reference factors for basic attribute tags include, in order, commonly used regions, age ranges, and gender information;
[0039] The reference factors for dietary habit labels include, in order of importance, taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences.
[0040] The reference factors for scenario demand tags include, in order, dining scenario, number of diners, real-time dietary needs, and dining time.
[0041] The reference factors for consumer behavior tags include, in order of importance, the average order value range, the frequency of ordering, the dishes that are repeatedly purchased, and the preference for dish pairings.
[0042] Compared with the prior art, the present invention has the following beneficial effects:
[0043] 1. This invention constructs a four-layer user profile tag using multi-dimensional data to accurately depict users' dietary preferences, scenario needs, consumption habits, and other characteristics, providing precise tag data support for subsequent dish selection and intent mining. Moreover, the dish selection is first based on the determination of the initial recommendation range, filtering out ingredients that are contraindicated or allergic, and finally matching the four layers of tags to select a basic dish pool, which improves the efficiency of subsequent matching.
[0044] 2. This invention employs a three-level intent determination logic through an intent mining unit. First, candidate intents are filtered through dietary habits and scenario demand tags. Then, basic attributes and consumption behavior tags are used for weighted optimization. Finally, real-time scenario data is used for dynamic correction, so that the dining intent determination not only meets the core needs of users but also adapts to the current real-time scenario, thereby enhancing the accuracy of menu recommendations. Attached Figure Description
[0045] Figure 1 This is a schematic diagram of a personalized menu push system based on user profiles according to the present invention. Detailed Implementation
[0046] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.
[0047] Example 1
[0048] like Figure 1 As shown, a personalized menu push system based on user profiles includes:
[0049] The data acquisition unit is used to collect multi-dimensional user data, which is obtained through parsing from the ordering system, self-filling pages, social media platform interfaces, and group-buying platform interfaces. As the data input source for the entire system, it collects multi-dimensional user data and outputs the pre-processed valid data to the user profiling unit, providing data support for the construction of user profile tags. Specifically, it acquires multi-dimensional user data through four channels: the ordering system, self-filling pages, social media platform interfaces, and group-buying platform interfaces. The ordering system acquires user ordering records, dish reviews, and consumption frequency; the self-filling pages acquire user basic information, dietary preferences, and actively entered data such as dietary restrictions / allergies; the social media platform interfaces acquire user food-related sharing and regional dietary preferences; and the group-buying platform interfaces acquire user group-buying consumption records, average order value range, and dining preferences. After collection, the data undergoes preliminary pre-processing to remove invalid and abnormal data, standardize the data format, and form a standardized multi-dimensional user dataset. Taking user Zhang as an example, the following data was collected from the ordering system: her ordering records for the past 30 days showed that she frequently ordered low-fat meals, repeatedly purchased Sichuan-style chicken breast bento boxes, had an average order value of 20-30 yuan, and disliked cilantro; data from the self-filled information page showed that she frequently visited Sichuan and Chongqing, was 20-30 years old, female, preferred mild spiciness, preferred chicken breast, and preferred low-fat light meals; data from the social media platform interface showed that she had shared content on "how to make Sichuan-style low-fat dishes," reflecting her preference for Sichuan-style low-fat dishes; and data from the group-buying platform interface showed that she had no group-buying records for group meals, only group-buying single-person light meals. After preprocessing the above data, invalid data with incorrect regional information was removed, and the data formats such as average order value and order frequency were standardized to form a standardized multi-dimensional dataset for Zhang.
[0050] The user profiling unit constructs user profile tags based on the multi-dimensional data. These tags consist of four layers: basic attribute tags, dietary habit tags, scenario-based needs tags, and consumption behavior tags. The dietary habit tags include multiple reference factors. Each tag in the user profiling unit also includes multiple reference factors: the basic attribute tags include frequently used regions, age ranges, and gender information; the dietary habit tags include taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences; the scenario-based needs tags include dining scenarios, number of diners, real-time dietary needs, and dining time; and the consumption behavior tags include average order value range, order frequency, repurchased dishes, and dish pairing preferences. During the construction of these four layers, the reference factors for each tag are quantified and categorized to form standardized tag data. Using Zhang as an example again, the constructed user profile would be:
[0051] Basic attribute tags: Commonly used region: Sichuan and Chongqing; Age range: 20-30 years old; Gender: Female;
[0052] Dietary habit tags: Flavor preference: mildly spicy; Food preference: chicken breast; Food allergies or dietary restrictions: cilantro; Diet type preference: low-fat light meals.
[0053] Scenario requirement tags: Overtime dining scenario, Single diners, Real-time convenient dining needs, Lunch during mealtime;
[0054] Consumer behavior tags: Average order value range 20-30 yuan, order frequency once a day, repeat purchase of Sichuan-style chicken breast bento, and preference for main dish + cold vegetable dish.
[0055] The dish selection unit includes a scenario determination module, used to set an initial recommendation range based on the basic attribute tags and the scenario requirement tags; a filtering module, used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; and a basic matching module, used to select matching dishes from the remaining dishes in the initial recommendation range based on the four layers of tags and add them to the basic dish pool. The dish selection unit comprises a scenario determination module, a filtering module, and a basic matching module, which are executed sequentially according to the process, performing layer-by-layer selection. The specific implementation method is as follows:
[0056] Scene Determination Module: Based on basic attribute tags and scene requirement tags, the initial recommended range of dishes is set. For example, the range of regionally compatible dishes is defined by combining the common regions of the basic attribute tags, and the range of single / multi-person dishes and convenient / group meals is defined by combining the dining scenario and number of diners of the scene requirement tags, forming an initial recommended dish set. This step can quickly narrow down the range of dishes and avoid a large number of mismatched dishes entering subsequent modules, which would increase the amount of useless data calculation. Based on Zhang's basic attribute tag "Sichuan and Chongqing" and scene requirement tags "overtime, single, lunch, convenient", the initial recommended range is defined as Sichuan-style single convenient light meals, including Sichuan-style chicken breast bento, low-fat fish bento, cucumber salad, unsweetened soy milk, etc., forming an initial recommended dish set.
[0057] Filtering module: Based on the reference factors of dietary habit tags for foods that users cannot eat or are allergic to, the initial recommended dish set is strictly filtered to remove all dishes containing foods that users cannot eat or are allergic to, forming a filtered dish set to ensure the safety of the recommended dishes. For example, Zhang's dietary habit tag indicates that he cannot eat cilantro, so dishes containing cilantro, such as "Sichuan-style cilantro chicken salad" and "cilantro tofu soup", are removed from the initial recommended dish set to form a filtered dish set.
[0058] Basic matching module: Based on the four-layer tags, the module filters out matching dishes from the filtered dish set and adds them to the basic dish pool. The specific method is as follows:
[0059] Calculate the matching scores of the dishes with each tag: basic attribute tag score S1, dietary habit tag score S2, scenario demand tag score S3, and consumption behavior tag score S4.
[0060] The weights for each label are preset as a, b, c, and d, respectively.
[0061] The overall matching score of the dishes is calculated using the following formula: S = a × S1 + b × S2 + c × S3 + d × S4;
[0062] A preset filtering threshold M is set. If the M ≤ S for a dish, then the dish is added to the basic dish pool.
[0063] Taking Mr. Zhang as a specific example, the matching scores of the filtered dishes, which are mainly Sichuan-style chicken breast bento boxes, are as follows: S1=90 points (highly suitable for Sichuan and Chongqing regions), S2=100 points (low-fat light meals, chicken breast as the main ingredient, mildly spicy flavor, no cilantro, perfectly matching dietary habits), S3=100 points (single person, overtime, convenient, lunch, perfectly matching the needs of the scenario), and S4=90 points (average order price of 20-30 yuan, repurchase category, highly matching consumer behavior). The preset weights a, b, c, and d are 10%, 40%, 30%, and 20% respectively, and the filtering threshold M=60 points is set. Then, the final comprehensive matching score S=90×10%+100×40%+100×30%+90×20%=97 points≥60 points, and it is added to the basic dish pool. Similarly, if dishes such as low-fat fish bento (S=92 points), cucumber salad (S=88 points), and unsweetened soy milk (S=90 points) all meet the threshold requirements, they are included in the basic dish pool; while the high-calorie Sichuan-style braised pork bento (S=45 points) is removed because it is below the threshold. This initial screening of all dishes by the dish selection unit avoids data redundancy during subsequent matching. Furthermore, this hierarchical screening method allows for interdependence between profile tags, providing a more comprehensive consideration and better suitability compared to the traditional method of matching profile tags to individual dishes.
[0064] The intent mining unit is used to acquire real-time scene data, including time period, weather, and user's current state, and to generate dining intent based on the four-layer tags and the real-time scene data. The intent mining unit includes a candidate intent module for acquiring candidate intents, the method for acquiring candidate intents being:
[0065] Establish a dining intent library, which includes multiple intent units, and the dining intents output by the intent mining unit can be composed of one or more intent units;
[0066] The dietary habit tag and scenario requirement tag in the four-layer label are associated with each intent unit, and the association matching score is calculated.
[0067] A preset candidate filtering threshold is used to set intent units with a correlation matching score greater than the threshold as candidate intents. Zhang's dietary habit tags (low-fat light meals, mildly spicy, preference for chicken breast, aversion to cilantro) and scenario demand tags (overtime, alone, lunch, convenience) are matched against the intent database. The calculated scores are: weight loss 85 points, convenient meals 80 points, stomach health 50 points, group meals 0 points, and satisfying cravings 60 points. The candidate filtering threshold is 70 points, therefore weight loss and convenient meals are set as candidate intents.
[0068] The intent mining unit also includes an auxiliary intent module, which calculates the matching score between the candidate intent and the basic attribute tag and consumption behavior tag in the four-layer tag. The auxiliary matching score is used as the auxiliary matching score. The association matching score and the auxiliary matching score are weighted and fused to obtain a comprehensive score. The candidate intent with the highest comprehensive score is selected as the core intent. In this embodiment, the association matching score weight is set to 70% and the auxiliary matching score weight is set to 30%. The comprehensive score of each candidate intent is calculated, and the candidate intent with the highest comprehensive score is selected as the core intent. If there are multiple candidate intents with similar comprehensive scores, the top two comprehensive scores can be set as the core intents. For example, Zhang's candidate intent and the auxiliary matching degree of basic attributes and consumption behavior tags: weight loss 90 points (25-year-old female from Sichuan and Chongqing, repurchase of low-fat meals, highly compatible with an average order value of 20-30 yuan), convenient dining 75 points (compatible with Sichuan and Chongqing region, single-person simple meal matching but repurchase category is low-fat meals); weighted and integrated calculation of comprehensive score: weight loss = 85×70%+90×30%=86.5 points, convenient dining = 80×70%+75×30%=78.5 points; weight loss was selected as the core intent.
[0069] The intent mining unit includes an intent fusion generation module, which calculates the matching degree by weighted fusion of core intents and real-time scene data, and obtains the core intent with the highest matching degree as the final dining intent. Similarly, the intent fusion generation module can also set a threshold to filter dining intents. If there are multiple core intents, the main core intent and auxiliary core intents can be determined according to the matching degree, which together form the final dining intent. For example, if Zhang's real-time scene data is obtained as lunchtime at 12:30, sunny day, and working overtime in the office, the matching degree of the core intent "weight loss" with this real-time scene is calculated to be 95 points (lunchtime is suitable for weight loss, working overtime requires low-fat and filling food, and sunny day does not affect the need for meals), therefore, Zhang's final dining intent is determined to be weight loss.
[0070] The intent mining unit includes a candidate intent module, an auxiliary intent module, and an intent fusion generation module. These three modules are executed sequentially according to the process to achieve three-level intent mining: "candidate intent screening - auxiliary score optimization - real-time scene fusion". The resulting dining intent is more tailored to the user.
[0071] A menu recommendation unit for selecting dishes from the basic dish pool according to the dining intention, combining them to generate a menu, and pushing it to the user. The menu recommendation unit includes a dish mapping module for converting the dining intention into quantifiable dish screening tags and establishing a mapping table between the dining intention and the dish tags, realizing the conversion from an abstract intention to a concrete product selection condition. The mapping table can be flexibly expanded according to the dining scenario and intention type. For example, converting the final dining intention of Zhang, which is "weight loss", into quantifiable dish screening tags, and establishing the mapping relationship as: weight loss → (fat ≤ 5g / 100g, protein ≤ 15g / 100g). With such quantitative indicators, the system can identify the dining intention corresponding to the dish.
[0072] A scenario feature module for establishing two types of scenarios, namely single-person set meals and multi-person set meals. The scenario includes the structure and quantity of the dishes. For example, the structure and quantity of the dishes are clearly defined in each scenario. The scenario feature of the single-person set meal is a fixed structure of "1 main course + 1 side dish + 1 drink", which is suitable for single-person dining intentions such as weight loss, stomach-nourishing, and convenient dining. The scenario feature of the multi-person set meal is a flexible structure of "multiple hot dishes + multiple cold dishes + 1 soup + drink combination", which is suitable for multi-person dining intentions such as dining together and satisfying cravings. The quantity of the dishes can be flexibly adjusted according to the number of diners. In this embodiment, since Zhang's dining intention is weight loss and the dining scenario is single-person overtime work, the single-person set meal scenario is matched, and the dish structure is "1 main course + 1 side dish + 1 drink".
[0073] A dish dynamic combination module for selecting dishes with a high degree of matching with the dining intention from the basic dish pool according to the mapping table and generating a combined set meal according to the scenario feature module. According to Zhang's weight loss dish screening tags (low fat, high protein, slightly spicy Sichuan flavor, single-person convenience, no coriander), select Sichuan-style chicken breast bento as the main course, cucumber salad as the side dish, and sugar-free soy milk as the drink from the basic dish pool, and combine them into a personalized weight loss set meal of "Sichuan-style chicken breast bento + cucumber salad + sugar-free soy milk" according to the single-person set meal structure, and push it to Zhang. At the same time, multiple replacement options can also be provided for him, such as replacing the main course with low-fat fish bento and replacing the side dish with wood ear salad, to meet the personalized selection needs of the user.
[0074] The menu recommendation unit includes a dish mapping module, a scenario feature module, and a dish dynamic combination module. The three modules work together to achieve a menu combination of "intention concretization - scenario standardization - set meal dynamization", which can facilitate users to directly select and improve the operation convenience.
[0075] The system also includes a personalized verification unit, used to periodically verify the four-layer tags in the user profile unit by setting an update time. As a closed-loop optimization unit of the system, it periodically verifies and updates the four-layer tags in the user profile unit to ensure the timeliness and accuracy of the tag data, thereby ensuring the accuracy of subsequent dish selection, intent mining, and menu recommendations. A fixed tag update time is preset, which is set to once a month in this embodiment. At the update time node, a comprehensive verification of the four-layer tags of the user profile unit is performed: combined with the multi-dimensional user data newly added by the data collection unit, such as: new order records, changes in consumption habits, addition of dietary restrictions, etc., the reference factors of each layer of tags are corrected, supplemented, or updated; tags that have not been used for a long time or are inconsistent with the latest user behavior are removed to ensure that the tag data is highly consistent with the current real situation of the user. Specific Case: On the 1st of each month, Zhang's four-layer tagging is verified. The data collection unit adds a record of him ordering low-fat fish bento boxes twice in the past 7 days. Therefore, the personalized verification unit updates his dietary habit tags, adding "fish" to the "ingredient preference" reference factors, while retaining "chicken breast"; the other tags remain unchanged and continue to be used. The updated tag data is synchronized to all downstream units. When recommending weight-loss meals to Zhang in the future, the low-fat fish bento box will be added as the preferred main dish option.
[0076] Example 2
[0077] A personalized menu push method based on user profiles includes the following steps:
[0078] Step S1, data collection, is used to collect multi-dimensional data of users. The multi-dimensional data is obtained through parsing the ordering system, the self-filling page, the social platform interface, and the group buying platform interface.
[0079] Step S2: Construct a user profile. Based on the multi-dimensional data, construct user profile tags. The user profile tags are four-layer tags, which include basic attribute tags, dietary habit tags, scenario demand tags, and consumption behavior tags in sequence. The dietary habit tags include multiple reference factors.
[0080] Step S3, dish selection, includes a scenario determination module, used to set an initial recommendation range based on the basic attribute tags and the scenario requirement tags; a filtering module, used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; and a basic matching module, used to select matching dishes from the remaining dishes in the initial recommendation range based on the four layers of tags and add them to the basic dish pool.
[0081] Step S4, Intent mining, obtaining real-time scene data, the real-time scene data including time period, weather and user current status, and generating dining intent based on the four-layer tags and the real-time scene data;
[0082] Step S5, Menu Recommendation: Based on the dining intention, select dishes from the basic menu pool, combine them to generate a menu, and push it to the user.
[0083] Each label in step S2 includes multiple reference factors, specifically:
[0084] The reference factors for basic attribute tags include, in order, commonly used regions, age ranges, and gender information;
[0085] The reference factors for dietary habit labels include, in order of importance, taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences.
[0086] The reference factors for scenario demand tags include, in order, dining scenario, number of diners, real-time dietary needs, and dining time.
[0087] The reference factors for consumer behavior tags include, in order of importance, the average order value range, the frequency of ordering, the dishes that are repeatedly purchased, and the preference for dish pairings.
[0088] Since Embodiment 1 and Embodiment 2 are essentially the same, the steps of the method in Embodiment 2 will not be described again.
[0089] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0090] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
Claims
1. A personalized menu push system based on user profiles, characterized in that, include: The data acquisition unit is used to collect multi-dimensional data from users, which is obtained by parsing the ordering system, self-filling page, social platform interface and group buying platform interface. The user profiling unit constructs user profile tags based on the multi-dimensional data. The user profile tags are four-layer tags, which include basic attribute tags, dietary habit tags, scenario demand tags, and consumption behavior tags. The dietary habit tags include multiple reference factors. The dish selection unit includes a scenario determination module, used to set an initial recommendation range based on the basic attribute tags and the scenario requirement tags; a filtering module, used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; and a basic matching module, used to select matching dishes from the remaining dishes in the initial recommendation range based on the four layers of tags and add them to the basic dish pool. The intent mining unit is used to acquire real-time scene data, which includes time period, weather and user current status, and to generate dining intent based on the four-layer tags and the real-time scene data. The menu recommendation unit is used to select dishes from the basic menu pool based on the user's dining intention, combine them to generate a menu, and push it to the user.
2. The personalized menu push system based on user profiles according to claim 1, characterized in that, Each tag in the user profile unit includes multiple reference factors, specifically: The reference factors for basic attribute tags include, in order, commonly used regions, age ranges, and gender information; The reference factors for dietary habit labels include, in order of importance, taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences. The reference factors for scenario demand tags include, in order, dining scenario, number of diners, real-time dietary needs, and dining time. The reference factors for consumer behavior tags include, in order of importance, the average order value range, the frequency of ordering, the dishes that are repeatedly purchased, and the preference for dish pairings.
3. The personalized menu push system based on user profiles according to claim 1, characterized in that, The method for adding matched dishes to the basic dish pool in the basic matching module includes: Calculate the matching scores of the dishes with each tag: basic attribute tag score S1, dietary habit tag score S2, scenario demand tag score S3, and consumption behavior tag score S4. The weights for each label are preset as a, b, c, and d, respectively. The overall matching score of the dishes is calculated using the following formula: S = a × S1 + b × S2 + c × S3 + d × S4; A preset filtering threshold M is set. If the M ≤ S for a dish, then the dish is added to the basic dish pool.
4. The personalized menu push system based on user profiles according to claim 1, characterized in that, The intent mining unit includes a candidate intent module, used to obtain candidate intents. The method for obtaining candidate intents is as follows: Establish a dining intent library, which includes multiple intent units, and the dining intents output by the intent mining unit can be composed of one or more intent units; The dietary habit tag and scenario requirement tag in the four-layer label are associated with each intent unit, and the association matching score is calculated. A preset candidate filtering threshold is set, and intent units with a correlation matching score greater than the candidate filtering threshold are set as candidate intents.
5. A personalized menu push system based on user profiles according to claim 4, characterized in that, The intent mining unit also includes an auxiliary intent module, which is used to calculate the matching score between the candidate intent and the basic attribute tag and consumption behavior tag in the four-layer tag. As an auxiliary matching score, the association matching score and the auxiliary matching score are weighted and fused to obtain a comprehensive score, and the candidate intent with the highest comprehensive score is selected as the core intent.
6. A personalized menu push system based on user profiles according to claim 5, characterized in that, The intent mining unit includes an intent fusion generation module, which is used to calculate the matching degree by weighted fusion of core intent and real-time scene data, and obtain the core intent with the highest matching degree as the final dining intent.
7. A personalized menu push system based on user profiles according to claim 1, characterized in that, The menu recommendation unit includes a dish mapping module, which is used to convert dining intentions into quantifiable dish filtering tags and establish a mapping table between dining intentions and dish tags. The scene feature module is used to create two types of scenes: single-person meal and multi-person meal. The scene includes the structure and quantity of dishes. The dynamic menu combination module is used to select dishes that match the dining intention from the basic menu pool according to the mapping table and generate a combination meal according to the scene feature module.
8. A personalized menu push system based on user profiles according to claim 1, characterized in that, The system also includes a personalized verification unit, which is used to set an update time to periodically verify the four-layer tags in the user profile unit.
9. A personalized menu push method based on user profiles, characterized in that, Includes the following steps: Step S1, data collection, is used to collect multi-dimensional data of users. The multi-dimensional data is obtained through parsing the ordering system, the self-filling page, the social platform interface, and the group buying platform interface. Step S2: Construct a user profile. Based on the multi-dimensional data, construct user profile tags. The user profile tags are four-layer tags, which include basic attribute tags, dietary habit tags, scenario demand tags, and consumption behavior tags in sequence. The dietary habit tags include multiple reference factors. Step S3, dish selection, including a scene determination module, is used to set an initial recommendation range based on the basic attribute tags and the scene requirement tags; The filtering module is used to remove dishes that are off-limits or allergenic from the initial recommendation range based on one of the reference factors in the dietary habit tags; the basic matching module is used to filter out matching dishes from the remaining dishes in the initial recommendation range based on the four-layer tags and add them to the basic dish pool. Step S4, Intent mining, obtaining real-time scene data, the real-time scene data including time period, weather and user current status, and generating dining intent based on the four-layer tags and the real-time scene data; Step S5, Menu Recommendation: Based on the dining intention, select dishes from the basic menu pool, combine them to generate a menu, and push it to the user.
10. A personalized menu push method based on user profiles according to claim 9, characterized in that, Each label in step S2 includes multiple reference factors, specifically: The reference factors for basic attribute tags include, in order, commonly used regions, age ranges, and gender information; The reference factors for dietary habit labels include, in order of importance, taste preferences, food preferences, foods to avoid or allergic to, and dietary type preferences. The reference factors for scenario demand tags include, in order, dining scenario, number of diners, real-time dietary needs, and dining time. The reference factors for consumer behavior tags include, in order of importance, the average order value range, the frequency of ordering, the dishes that are repeatedly purchased, and the preference for dish pairings.