An AI recommendation ordering method, system and electronic device

Abstract

The application relates to an AI recommendation ordering method, system and electronic equipment, wherein the method comprises the following steps: extracting at least two user intents from a real-time acquired natural language text; determining a priority weight of the user intents based on a context and a historical behavior of a user; determining a target model from a pre-acquired entity extraction model based on the priority weight, and extracting entity information in the corresponding user intent through the target model, wherein different priority weights correspond to different target models; and sending a recommended card of a candidate commodity to the user based on the entity information, wherein the recommended card comprises a purchase interface of the candidate commodity, so that the user can complete a purchase operation.

Description

Technical Field

[0001] This application relates to the field of AI-recommended order placement, and in particular to an AI-recommended order placement method, system, and electronic device. Background Technology

[0002] With the rise and widespread application of AI, users have increasingly higher demands for AI services. However, existing AI systems, when users have purchasing needs, are unable to dynamically distinguish between core and secondary needs from the natural language text input by the user, resulting in an inability to accurately identify user requirements.

[0003] Therefore, existing technologies fail to accurately recommend the products users need due to inaccurate intent recognition, thus affecting user experience. Summary of the Invention

[0004] This application provides an AI-powered order recommendation method, system, and electronic device to at least address the problem in related technologies where inaccurate intent recognition leads to an inability to accurately recommend desired products to users, thus affecting user experience.

[0005] Firstly, embodiments of this application provide an AI-recommended order placement method, including:

[0006] Extract at least two user intents from real-time acquired natural language text, and determine the priority weight of the user intents based on context and the user's historical behavior;

[0007] Based on the priority weights, a target model is determined from the pre-acquired entity extraction model, and entity information corresponding to the user intent is extracted through the target model. Different priority weights correspond to different target models.

[0008] Based on the entity information, a recommendation card for candidate products is sent to the user. The recommendation card includes a purchase interface for the candidate products, allowing the user to complete the purchase.

[0009] In one embodiment, determining the priority weight of the user intent based on context and the user's historical behavior includes:

[0010] The urgency weight corresponding to each user intent is determined based on the context and the first keyword in each user intent;

[0011] In response to any of the emergency weights being greater than or equal to the first weight threshold, the emergency weight is used as the priority weight.

[0012] In one embodiment, determining the priority weight of the user intent based on context and the user's historical behavior further includes:

[0013] In response to each of the urgency weights being less than the first weight threshold, based on the context and the user's historical behavior, an explicit weight and a context-related weight corresponding to the user's intent are determined;

[0014] The priority weight corresponding to the user intent is determined based on the explicit weight, contextual weight, and urgency weight.

[0015] In one embodiment, determining the explicit weight and context-related weight corresponding to the user intent based on the context and the user's historical behavior includes:

[0016] The explicit weight of the user intent is determined based on the second keyword and semantic structure in the user intent;

[0017] The contextual association weight of the user's intent is determined based on the context and the user's historical behavior.

[0018] In one embodiment, determining the target model from the pre-acquired entity extraction model based on the priority weights includes:

[0019] In response to the priority weight being greater than or equal to the second weight threshold, the first model is selected as the target model;

[0020] In response to the priority weight being less than the second weight threshold, the second model is selected as the target model, and the model size and computational cost of the second model are less than those of the first model.

[0021] In one embodiment, the method further includes:

[0022] If the explicit weight of the user's intent is less than the third weight threshold, an intent confirmation question is provided to the user.

[0023] In response to the answer to the user's input intent confirmation question, the user intent is re-determined based on the answer and the natural language text.

[0024] In one embodiment, sending a recommendation card for candidate products to the user based on the entity information includes:

[0025] Acquire environmental data and initial user behavior data, and determine the user's first shopping personality based on the environmental data, the behavior data, and a pre-constructed feature-personality-product knowledge graph;

[0026] Based on the entity information, a first recommendation strategy corresponding to the first shopping personality is obtained from the feature-personality-product knowledge graph, and product recommendations are made based on the first recommendation strategy.

[0027] Obtain user behavior data based on recommended product feedback, and determine the user's second shopping personality based on the environmental data, the behavior data, and the feature-personality-product knowledge graph;

[0028] Based on the entity information, a second recommendation strategy corresponding to the second shopping personality is obtained from the feature-personality-product knowledge graph, and products are recommended based on the second recommendation strategy.

[0029] In one embodiment, sending a recommendation card for candidate products to the user based on the entity information includes:

[0030] The recommended card type is determined from the first mapping relationship based on the type of the candidate product, wherein the first mapping relationship is a pre-stored mapping relationship between the candidate product type and the recommended card type;

[0031] Recommended card information is determined from the second mapping relationship based on the recommended card type. The second mapping relationship is a pre-stored mapping relationship between recommended card types and recommended card information.

[0032] Secondly, embodiments of this application provide an AI-recommended order placement system, including:

[0033] Weight determination module: used to extract at least two user intents from real-time acquired natural language text, and determine the priority weight of the user intents based on context and the user's historical behavior;

[0034] Entity extraction module: used to determine a target model from the pre-acquired entity extraction model based on the priority weights, and extract entity information corresponding to the user intent through the target model, wherein different priority weights correspond to different target models;

[0035] Recommendation module: Used to send recommendation cards of candidate products to users based on the entity information. The recommendation card includes the purchase interface of the candidate products so that users can complete the purchase operation.

[0036] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the AI ​​recommendation ordering method as described in the first aspect above.

[0037] The AI-recommended order placement method, system, and electronic device provided in this application embodiment have at least the following technical effects.

[0038] This application dynamically adjusts the weight of user intent based on context and historical behavior, which is more conducive to accurately identifying the user's primary and secondary intents, thereby meeting user needs based on precise intents and improving user experience. After determining the priority weight of user intents, different target models are used for entity extraction for different priority weights. While ensuring the accuracy of entity extraction, the model size and computational load are reduced, effectively saving storage and computing resources. Based on entity information, product purchase interfaces are directly recommended to users, eliminating the need for additional redundant operations and enabling "one-click" ordering, thus improving user experience. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0040] Figure 1 This is a flowchart illustrating an AI-recommended order placement method according to an embodiment of this application;

[0041] Figure 2 This is a structural block diagram of an AI recommendation ordering system according to an embodiment of this application;

[0042] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this application clearer, the application is described and illustrated below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. All other embodiments obtained by those skilled in the art based on the embodiments provided in this application without inventive effort are within the scope of protection of this application.

[0044] Obviously, the accompanying drawings described below are merely some examples or embodiments of this application. Those skilled in the art can apply this application to other similar scenarios based on these drawings without any inventive effort. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this application, any changes to design, manufacturing, or production based on the technical content disclosed in this application are merely conventional technical means and should not be construed as insufficient disclosure of the content of this application.

[0045] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment that is mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments without conflict.

[0046] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0047] Firstly, embodiments of this application provide an AI-recommended order placement method. This method is applied to an AI-recommended order placement system, which is presented as software or a webpage. Unlike traditional shopping scenarios, this AI-recommended order placement method is specifically applied to conversational order placement scenarios, receiving natural language text input by the user containing the user's intent, and pushing target products to the customer based on the natural language text.

[0048] Figure 1 This is a flowchart illustrating an AI-recommended order placement method according to an embodiment of this application, such as... Figure 1 As shown, the method includes:

[0049] Step S101: Extract at least two user intents from the real-time acquired natural language text, and determine the priority weight of the user intents based on the context and the user's historical behavior.

[0050] Optionally, the real-time acquired natural language text includes, but is not limited to, at least one of the following: text input by the user, text converted from the user's voice input, and text extracted from the user's image input. User intent is differentiated based on historical behavior, context, and keywords, and user priority weights are determined. Contextual information is used to determine the direction of intent based on the user's current historical dialogue content; historical behavior includes the user's historical preferences and behavioral models, used to weight intents. In this way, by dynamically adjusting intent weights based on context and historical behavior, it is more conducive to accurately identifying the user's primary and secondary intents, thereby meeting user needs based on precise intents and improving user experience.

[0051] In one example, step S101 includes:

[0052] Step S1011: Determine the urgency weight of the corresponding user intent based on the context and the first keyword in each user intent.

[0053] Step S1012: In response to any emergency weight being greater than or equal to the first weight threshold, the emergency weight is used as the priority weight.

[0054] Optionally, the urgency weight represents the urgency of the user's need; the context represents the current user's historical conversation content; the primary keyword is a time-sensitive word, such as "immediately" or "right away," indicating that the user urgently needs a certain service. If a time-sensitive word appears, it indicates that the user's need is more urgent, and the corresponding urgency weight value increases. When the urgency weight value is greater than or equal to a preset threshold, the priority of the user's intent is dynamically adjusted directly through the urgency weight. In this way, timely and accurate responses to users' urgent needs are provided, improving the user experience.

[0055] In one example, step S101 further includes:

[0056] Step S1013: In response to each emergency weight being less than the first weight threshold, determine the explicit weight and context-related weight of the corresponding user intent based on the context and the user's historical behavior.

[0057] Step S1014: Determine the priority weight of the corresponding user intent based on explicit weight, contextual weight, and urgency weight.

[0058] Optionally, when the urgency weight is less than a preset threshold, the priority of the user intent is dynamically adjusted based on the sum of the explicit weight, contextual weight, and urgency weight. The explicit weight indicates whether the user intent is direct and clear; the clearer the intent, the higher the explicit weight. The contextual weight indicates the degree to which contextual information supplements the intent; if the contextual information includes intent weight-related information, the corresponding contextual weight is higher. This method of dynamically adjusting intent weights based on context and historical behavior information is more conducive to accurately identifying the user's primary and secondary intents, thereby meeting user needs based on precise intents and improving user experience.

[0059] In one example, step S1013 includes:

[0060] Step S301: Determine the explicit weight of the user intent based on the second keyword and semantic structure in the user intent.

[0061] Optionally, explicit weights characterize whether the user's intent is direct and clear; the clearer the user's intent, the higher the explicit weight. The second keyword refers to keywords that clearly identify the user entity, such as core action words and object entity words in the user's intent, like "purchase," "package," and "headphones." Semantic structure refers to the semantic connections and logical organization between language units (words, phrases, sentences), such as subject-verb-object-result structures, parallel selection structures, etc. Generally, the explicit weight of subject-verb-object structures may be higher. As an example, suppose a user inputs "buy airline tickets or high-speed rail tickets," indicating multiple purchase options. The explicit weight of the user's intent is not high; the explicit weight of "buy airline tickets" is the same as that of "buy high-speed rail tickets."

[0062] Step S302: Determine the contextual association weight of the user's intent based on the context and the user's historical behavior.

[0063] Optionally, the context represents the current user's historical dialogue content, used to supplement and weight the user's intent. For example, if a user's request via natural language input is "buy a Nezha movie ticket this afternoon," and the context mentions related topics such as "play," "entertainment," and "weekend," the contextual weight of the user's intent will increase. Historical behavior includes the user's historical browsing and purchasing actions, such as the duration of the user's stay on the current page and the user's purchase history. For example, assuming the user's historical behavior shows that they always order a movie whenever they mention entertainment, the contextual weight of the user's intent will be increased based on this historical behavior. In this way, contextual information and historical behavior information are effectively utilized to improve dialogue state management capabilities. When user needs change or are added, there is no need to repeatedly input key information, thus improving the user experience.

[0064] As an example, suppose a user inputs: "I want to order a lunch set meal for one person, the most affordable one." In this scenario, the extracted user intents include: order a lunch set meal, most affordable, and for one person. Calculate the urgency weight of each intent.

[0065] meaning Figure 1 Order a lunch set meal. Urgency weight: 7 / 10 ("Ordering one" indicates an immediate need, thus indicating a high degree of time urgency).

[0066] meaning Figure 2 Best deal, urgency weight: 6 / 10 (relatively urgent, but not as important as lunch set meal).

[0067] meaning Figure 3 : Single person, urgency weight: 2 / 10 (single person does not affect time urgency).

[0068] In this scenario, if the first weight threshold is 7, the priority of the intent can be directly adjusted based on the urgency weight. If the first weight threshold is 8, all intents are less than the first weight threshold, and the priority weight of the corresponding user intent needs to be determined based on the explicit weight, contextual weight, and urgency weight. It should be noted that the first weight threshold can be set according to the actual application scenario and is not limited to the examples listed in this application. When the first weight threshold is 8, the priority weight of each user intent is calculated.

[0069] meaning Figure 1 : Order a lunch set meal. Explicit weight: 8 / 10 (user explicitly states the need); Contextual weight: 3 / 10 (if the user has made a similar request before, the weight will be higher); Urgency weight: 7 / 10 ("Ordering one" indicates an immediate need, thus indicating a high degree of time urgency); Final priority weight: 18 / 30.

[0070] meaning Figure 2 : Best deal. Explicit weight: 7 / 10 (User specifies price range, select the highest discount package according to system discount calculation rules); Contextual weight: 0 / 10 (No historical preference); Urgency weight: 6 / 10 (Relatively urgent, but not as important as lunch package); Final priority weight: 13 / 30.

[0071] meaning Figure 3 : Single person. Explicit weight: 5 / 10 (portion mentioned but not a core requirement); Contextual weight: 4 / 10 (if the user has inquired about a single dish); Urgency weight: 2 / 10 (single person does not affect time urgency); Final priority weight: 11 / 30.

[0072] Calculate the priority weight of each intent, dynamically determine the priority order of intents, and identify the primary intent (order a lunch set meal) and secondary intents (best deal, single person).

[0073] Step S102: Determine the target model from the pre-acquired entity extraction model based on priority weights, and extract entity information corresponding to the user intent through the target model. Different priority weights correspond to different target models.

[0074] Optionally, different priority weights correspond to different entity extraction models. For intents with higher priority weights, entities are extracted using a high-precision model; for intents with lower priority weights, entities are extracted using a lightweight model. In this way, entities from high-priority intents are extracted using a high-precision model, ensuring accuracy, while entities from low-priority intents are extracted using a lightweight model, reducing the computational cost of entity extraction. This reduces model size and computational cost while maintaining entity extraction accuracy, effectively saving storage and computing resources.

[0075] In one example, step S102 includes:

[0076] Step S1021: In response to the priority weight being greater than or equal to the second weight threshold, the first model is selected as the target model.

[0077] In step S1022, in response to the priority weight being less than the second weight threshold, the second model is selected as the target model. The model size and computational cost of the second model are less than those of the first model.

[0078] Optionally, the first model is a high-precision model, such as the BERT model combined with a conditional random field, or the RoBERTa-large model combined with a conditional random field. The second model is a lightweight model, such as a lightweight BiLSTM model or a LiteTransformer. The first model has higher accuracy, but the high-precision model has a larger model size and higher computational cost. By using different entity extraction models to extract entities from user intent, the system ensures accurate extraction of entities representing the primary intent while reducing the computational cost of extracting secondary intents. This reduces model size and computational cost while maintaining entity extraction accuracy, effectively saving storage and computational resources.

[0079] Step S103: Based on the entity information, send a recommendation card for the candidate products to the user. The recommendation card includes the purchase interface for the candidate products so that the user can complete the purchase operation.

[0080] Optionally, product cards are recommended to users based on entity information. These product cards include purchase interfaces for candidate products and links to the products, allowing users to directly purchase the target product via the link. A single product card can contain more than one product. For example, when a customer orders takeout, a product card pushed to the customer can include multiple meal deals from various stores. Existing technologies, after recognizing user intent, can only display product names or categories, requiring users to manually search and purchase the recommended products through relevant software, leading to process redundancy and error accumulation. This application directly recommends product cards to users, eliminating the need for additional redundant operations and enabling "one-click" ordering, thus improving the user experience.

[0081] In one example, step S103 includes:

[0082] The recommended card type is determined from the first mapping relationship based on the type of the candidate product. The first mapping relationship is a pre-stored mapping relationship between candidate product types and recommended card types.

[0083] Recommended card information is determined from the second mapping relationship based on the recommended card type. The second mapping relationship is a pre-stored mapping relationship between recommended card types and recommended card information.

[0084] Optionally, the card type can be determined based on the type of candidate product. Different card types correspond to different card information, and the corresponding card information is determined according to the card type. For example, the card information corresponding to product types such as group-buying products, food ordering products, and recharge products is different. As an example, the card information for recharge products includes information such as the recharge store, recharge type, amount, and recharge link. The card information for group-buying products includes information such as the group-buying store, group-buying product and details, and purchase link. Food ordering products include information such as store information, store menu, order link, and estimated delivery time.

[0085] This example demonstrates how recommending cards are sent to users based on different product types, showcasing relevant information about the recommended products so users can directly view and purchase them. This simplifies user operations and improves the user experience.

[0086] In one example, the method also includes:

[0087] If the explicit weight of the user's intent is less than the third weight threshold, an intent confirmation question is provided to the user.

[0088] In response to the answer to the user's input intent confirmation question, the user's intent is re-determined based on the answer and natural language text.

[0089] Optionally, for cases with ambiguous intent, such as a user simply saying "What's good to eat?", the explicit weight of the corresponding user intent is relatively low. In this case, guiding questions can be designed to further clarify the user intent. For example, "Do you want to buy a group-buying deal, or order food / takeout directly?" After the user replies, the user intent is redefined based on their response. Dynamic priorities, entity extraction, and product recommendations are then performed based on the redefined intent. If the user does not reply, the option of both services can be automatically provided, such as "Here are some group-buying deals; you can also order food directly."

[0090] As an example, Table 1 is a table of factors for judging ordering intention.

[0091] Table 1

[0092]

[0093] As shown in Table 1, user intent is accurately determined based on historical user behavior, keyword features, and semantic parsing. For example, if a user's sentence contains related words such as "group buying" and "coupons," the intent leans towards "buying group buying coupons." The weight of user context association can be adjusted based on historical preferences, such as P(group buying coupons | keyword matching + access path + past consumption records); P(ordering food | keyword matching + access path + past consumption records). When the explicit weight of user intent is low, guiding questions are used to further clarify the user intent.

[0094] In summary, this application dynamically adjusts intent weights based on context and historical behavior information, which is more conducive to accurately identifying the user's primary and secondary intents, thereby meeting user needs based on precise intents and improving user experience. Furthermore, by effectively utilizing contextual and historical behavior information, it improves dialogue state management capabilities, eliminating the need to repeatedly input key information when user needs change or are added, further enhancing user experience. After determining the priority weights of user intents, a high-precision model extracts entities of high-priority intents to ensure accuracy, while a lightweight model extracts entities of low-priority intents to reduce computational load. This reduces model size and computational load while maintaining entity extraction accuracy, effectively saving storage and computing resources. Finally, this application directly recommends product cards to users based on entity information, eliminating the need for redundant user operations and enabling "one-click" ordering, further improving user experience.

[0095] In some embodiments, the purchase interface for step S103, which recommends candidate products to the user based on entity information, includes: classifying user intent, distinguishing between purchase intent, price comparison intent, and inquiry intent, etc.; retrieving products from the database based on entity information; calculating product ratings based on entity matching degree, user historical behavior, product popularity, and other information; and recommending products with higher ratings to the user.

[0096] Preferably, step S103 is implemented by the following steps:

[0097] Step S1031: Obtain environmental data and initial user behavior data. Based on the environmental data, behavior data, and pre-built feature-personality-product knowledge graph, determine the user's primary shopping personality.

[0098] Environmental data includes, but is not limited to, time (e.g., holidays, weekdays), geographical location (e.g., office area, residential area), device type (e.g., mobile device, PC device), weather, social trends (e.g., trending products), and inventory status. User behavioral data includes, but is not limited to, multimodal input information (e.g., text input, voice input, image input), click frequency, page scrolling speed, and price comparison behavior. It should be noted that the initial behavioral data is usually the input content. By extracting features from the environmental data and the initial behavioral data, the extracted features are matched with a feature-personality-product knowledge graph to confirm the user's primary shopping personality.

[0099] Step S1032: Based on entity information, obtain the first recommendation strategy corresponding to the first shopping personality from the feature-personality-product knowledge graph, and recommend products based on the first recommendation strategy.

[0100] The feature-personality-product knowledge graph includes the mapping relationship between feature parameters and shopping personality, as well as the mapping relationship between shopping personality and product recommendation strategy. Entity information, as a type of feature information, is used to determine the shopping personality and make product recommendations. Table 2 is an example table of mapping relationships for a feature-personality-product knowledge graph according to an embodiment of this application.

[0101] Table 2

[0102]

[0103] Step S1033: Obtain user behavior data based on recommended product feedback, and determine the user's second shopping personality based on environmental data, behavioral data, and feature-personality-product knowledge graph.

[0104] The system recommends products based on the primary recommendation strategy, then acquires real-time user feedback behavior data to determine if the user's shopping personality has changed. If so, it updates the shopping personality in real-time using lightweight edge computing, ensuring a seamless switch for the user (e.g., switching from "efficiency type" to "stay-at-home dad type" requires only a 300ms response delay). It's important to note that the primary and secondary shopping personalities can be the same. That is, if the shopping personality determined by user feedback behavior data analysis remains the user's primary shopping personality, product recommendations continue to be made based on that primary personality.

[0105] The feature-personality-product knowledge graph stores the personality weight of each shopping personality and the feature weights of feature parameters under each shopping personality. In steps S1031 and S1033, based on environmental data, behavioral data, and the pre-constructed feature-personality-product knowledge graph, the user's shopping personality is determined. The shopping personality includes a first shopping personality and a second shopping personality.

[0106] Step S201: Extract target feature parameters from environmental data and behavioral data.

[0107] In some embodiments, step S201 includes:

[0108] Step S2011: Extract time parameters, meteorological parameters, location parameters, and device type parameters from the environmental data. The time parameters include date and time parameters, and the meteorological parameters include weather and seasonal parameters. Optionally, the date parameter determines whether it is a weekday or holiday, the clock parameter determines whether it is morning, noon, or evening, the location parameter determines whether the user is in an office area or a residential area, and the device parameter determines whether the user is using a mobile device or a PC.

[0109] Step S2012: Extract active input parameters and browsing behavior parameters from the behavioral data. Active input parameters include keyword content parameters (e.g., product category, brand, product features) and keyword specificity parameters; browsing behavior parameters include, but are not limited to, click frequency, page scrolling speed, price comparison behavior, and cross-category navigation paths. Optionally, the extraction rule for keyword specificity parameters is that inputting "brand + precise product name" (e.g., brand name + Americano coffee) is considered a specific keyword, while inputting only "product category" (e.g., coffee) is considered an unclear keyword.

[0110] Step S202: Calculate the feature value of each shopping personality based on the feature weights of the target feature parameters under each shopping personality.

[0111] For example, the target feature parameters include feature a, feature b, and feature c, and the initial value of each feature is set to 1. The feature parameter corresponding to Shopping Personality I includes feature a, and the feature weight of feature a is 40%, so the feature value of Shopping Personality I is 0.4; the feature parameters corresponding to Shopping Personality II include features a and feature b, and the feature weight of feature a is 30% and the feature weight of feature b is 20%, so the feature value of Shopping Personality II is 0.3 + 0.2 = 0.5; the feature parameter corresponding to Shopping Personality III includes feature c, and the feature weight of feature c is 60%, so the feature value of Shopping Personality III is 0.6.

[0112] Step S203: Based on the characteristic values ​​and personality weights of each shopping personality, determine the evaluation value for each shopping personality, and select the shopping personality with the highest evaluation value as the user's shopping personality. It should be noted that the personality weights are determined based on historical behavioral data. Depending on the actual situation, the evaluation value can be the product of the characteristic value and the personality weight, or it can be a weighted sum.

[0113] Taking the aforementioned Shopping Personality I, Shopping Personality II, and Shopping Personality III as examples. Assume that the personality weight of Shopping Personality I is 10%, the personality weight of Shopping Personality II is 15%, and the personality weight of Shopping Personality III is 5%, and the evaluation value = trait value * personality weight. The evaluation value of Shopping Personality I is 0.4 * 10% = 0.04, the evaluation value of Shopping Personality II is 0.5 * 15% = 0.075, and the evaluation value of Shopping Personality III is 0.6 * 5% = 0.03. Shopping Personality III has the highest evaluation value; therefore, Shopping Personality III is taken as the user's current shopping personality.

[0114] When multiple assessment values ​​are close, the shopping personality corresponding to the stronger scenario signal is responded to first according to the preset signal priority. For example, if the signal priority is input signal > browsing signal > environmental signal, and the assessment values ​​of the efficiency type and the emergency type are the same, and the AI ​​dialogue search contains the keyword "urgent", then the emergency type personality will be responded to first.

[0115] This embodiment determines feature values ​​through multi-source real-time data (current environmental data and behavioral data), determines personality weights through historical data, and obtains the evaluation value of shopping personality based on feature values ​​and personality weights. In the process of personality analysis, multi-source real-time data and historical data are considered at the same time to improve the accuracy of the analysis results, so that the shopping personality obtained by the analysis is consistent with the user's current shopping psychology, thereby improving recommendation efficiency.

[0116] Step S1034: Based on entity information, obtain the second recommendation strategy corresponding to the second shopping personality from the feature-personality-product knowledge graph, and recommend products based on the second recommendation strategy.

[0117] When a user's shopping personality changes, a new shopping personality (second shopping personality) needs to be switched for product recommendations. For example, if analysis shows that the user's first shopping personality is efficiency-oriented, and product recommendations are made based on this personality, but then the user is detected frequently browsing mother and baby products, this behavior analysis indicates that the user has shifted to a "dad" personality. Therefore, the current shopping personality will be switched to the "dad" personality for product recommendations.

[0118] Through the above steps S1031 to S1034, environmental data and user behavior data are analyzed in real time. Based on the analysis results, the shopping personality is adaptively switched to dynamically adapt to user needs, thereby improving the accuracy and efficiency of product recommendations and solving the problem of low product recommendation efficiency.

[0119] In some embodiments, recommendation strategies corresponding to shopping personalities are obtained from a feature-personality-product knowledge graph. The shopping personalities include a first shopping personality and a second shopping personality. The recommendation strategies include a first recommendation strategy and a second recommendation strategy.

[0120] When the shopping personality is efficiency-oriented, the efficiency-oriented personality is matched with the feature-personality-product knowledge graph to obtain a recommendation strategy that recommends products from high to low sales volume and displays a parameter comparison bar.

[0121] When the shopping personality is exploratory, the exploratory personality is matched with the trait-personality-product knowledge graph to obtain a recommendation strategy that prioritizes new products, seasonal limited products, and less popular products.

[0122] Shopping personality can be set according to actual situation, including but not limited to "efficiency type", "exploration type", "parenting type", "emergency type", "entertainment type" and "economic type".

[0123] In some embodiments, the method further includes:

[0124] S106 After a user purchases a product, the weight data in the feature-personality-product knowledge graph is adjusted based on the current purchase behavior data. The weight data includes personality weight and feature weight.

[0125] By optimizing the personality weight and adjusting the weight in the feature-personality-product knowledge graph through purchase behavior feedback, for example, if a user makes a purchase after making a high-frequency parameter comparison in the efficiency personality state, the weight of the efficiency personality will be increased. At the same time, the weight of the "parameter comparison" feature in the efficiency personality will be increased.

[0126] S107, based on the time decay function, adjusts the weight data in the feature-personality-product knowledge graph. The weight data includes personality weight and feature weight.

[0127] A time decay function is introduced to reduce the weight of historical behavior, ensuring that the knowledge graph reflects the user's latest preferences. For example, a product purchased a year ago should have less impact on current recommendations than a recently purchased product.

[0128] In some embodiments, the method further includes:

[0129] S108, based on user behavior data based on feedback from recommended products, adjust the weight data of the corresponding node of the recommended product in the feature-personality-product knowledge graph. The node includes shopping personality and feature parameters, and the weight data includes personality weight and feature weight.

[0130] Based on user feedback on the recommendations (e.g., clicking to view, ignoring, or disliking), the relevant weight data in the knowledge graph is adjusted. Optionally, if a user repeatedly ignores a certain type of recommended product, the weight of the corresponding personality or trait in the knowledge graph is reduced. For example, if a user continuously ignores movie recommendations, the probability of triggering the "entertainment-type" personality trait is reduced.

[0131] By optimizing the personality weight and adjustment weight in the feature-personality-product knowledge graph through a feedback mechanism, we can further ensure that the recommended products match the user's shopping psychology and improve recommendation efficiency.

[0132] Secondly, embodiments of this application provide an AI-recommended order placement system. Figure 2 This is a structural block diagram of an AI recommendation ordering system according to an embodiment of this application, such as... Figure 2 As shown, the system includes:

[0133] Weight determination module 100: used to extract at least two user intents from real-time acquired natural language text, and determine the priority weight of the user intents based on context and the user's historical behavior.

[0134] Entity extraction module 200: used to determine the target model from the pre-acquired entity extraction model based on priority weights, and extract entity information corresponding to the user intent through the target model. Different priority weights correspond to different target models.

[0135] Recommendation module 300: Used to send recommendation cards of candidate products to users based on entity information. The recommendation cards include the purchase interface of the candidate products so that users can complete the purchase operation.

[0136] In one example, the weight determination module 100 includes:

[0137] This is used to determine the urgency weight of the corresponding user intent based on the context and the first keyword in each user intent.

[0138] In response to any emergency weight being greater than or equal to the first weight threshold, the emergency weight is used as the priority weight.

[0139] In one example, the weight determination module 100 also includes:

[0140] In response to each emergency weight being less than a first weight threshold, the explicit weight and context-related weight of the corresponding user intent are determined based on the context and the user's historical behavior.

[0141] Priority weights are used to determine the corresponding user intent based on explicit weights, contextual weights, and urgency weights.

[0142] In one example, based on context and the user's historical behavior, the explicit weight and context-related weight of the corresponding user intent are determined, including:

[0143] The explicit weight of user intent is determined based on the second keyword and semantic structure in the user intent.

[0144] Contextual weights are used to determine user intent based on context and the user's historical behavior.

[0145] In one example, entity extraction module 200 includes:

[0146] If the priority weight is greater than or equal to the second weight threshold, the first model is selected as the target model.

[0147] In response to the priority weight being less than the second weight threshold, the second model is selected as the target model. The model size and computational cost of the second model are less than those of the first model.

[0148] In one example, the system also includes:

[0149] If the explicit weight of the user's intent is less than the third weight threshold, an intent confirmation question is provided to the user.

[0150] In response to the answer to the user's input intent confirmation question, the user's intent is re-determined based on the answer and natural language text.

[0151] In one example, recommendation module 300 includes:

[0152] Used to acquire environmental data and initial user behavior data, and based on the environmental data, behavior data and a pre-built feature-personality-product knowledge graph, to determine the user's primary shopping personality.

[0153] Based on entity information, the first recommendation strategy corresponding to the first shopping personality is obtained from the feature-personality-product knowledge graph, and product recommendations are made based on the first recommendation strategy.

[0154] By acquiring user behavior data based on product recommendations, and using environmental data, behavioral data, and a feature-personality-product knowledge graph, we can determine the user's secondary shopping personality.

[0155] Based on entity information, a second recommendation strategy corresponding to the second shopping personality is obtained from the feature-personality-product knowledge graph, and products are recommended based on the second recommendation strategy.

[0156] In one example, recommendation module 300 includes:

[0157] The recommended card type is determined from the first mapping relationship based on the type of the candidate product. The first mapping relationship is a pre-stored mapping relationship between candidate product types and recommended card types.

[0158] Recommended card information is determined from the second mapping relationship based on the recommended card type. The second mapping relationship is a pre-stored mapping relationship between recommended card types and recommended card information.

[0159] In summary, this application utilizes a weight determination module to dynamically adjust intent weights based on context and historical behavior information. This approach is more effective in accurately identifying the user's primary and secondary intents, thereby meeting user needs based on precise intents and improving user experience. Furthermore, by effectively utilizing contextual and historical behavior information, it enhances dialogue state management capabilities. When user needs change or are added, there is no need to repeatedly input key information, further improving user experience. After determining the priority weights of user intents, the entity extraction module uses a high-precision model to extract entities for high-priority intents, ensuring accuracy. A lightweight model extracts entities for low-priority intents, reducing computational overhead. This reduces model size and computational load while maintaining entity extraction accuracy, effectively saving storage and computing resources. The recommendation module directly recommends product cards to users based on entity information, eliminating the need for redundant user operations and enabling "one-click" ordering, further improving user experience.

[0160] Thirdly, embodiments of this application provide an electronic device, Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the AI ​​recommendation order placement method provided in the first aspect. Figure 3 The electronic device 60 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0161] Electronic device 60 may be in the form of a general-purpose computing device, such as a server device. Components of electronic device 60 may include, but are not limited to: at least one processor 61, at least one memory 62, and a bus 63 connecting different system components (including memory 62 and processor 61).

[0162] Bus 63 includes a data bus, an address bus, and a control bus.

[0163] The memory 62 may include volatile memory, such as random access memory (RAM) 621 and / or cache memory 622, and may further include read-only memory (ROM) 623.

[0164] The memory 62 may also include a program / utility 625 having a set (at least one) of program modules 624, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0165] The processor 61 executes various functional applications and data processing by running computer programs stored in the memory 62, such as the AI ​​recommendation order placement method provided in the first aspect of this application.

[0166] Electronic device 60 can also communicate with one or more external devices 64 (e.g., keyboard, pointing device, etc.). This communication can be performed via input / output (I / O) interface 65. Furthermore, the model-generated electronic device can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public network, such as the Internet) via network adapter 66. As shown, network adapter 66 communicates with other modules of the model-generated electronic device via bus 63. It should be understood that, although not shown in the figure, other hardware and / or software modules can be used in conjunction with the model-generated electronic device, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID (disk array) systems, tape drives, and data backup storage systems.

[0167] It should be noted that although several units / modules or sub-units / modules of the electronic device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of the present invention, the features and functions of two or more units / modules described above can be embodied in one unit / module. Conversely, the features and functions of one unit / module described above can be further divided and embodied by multiple units / modules.

[0168] Fourthly, embodiments of this application provide a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the AI ​​recommendation order placement method provided in the first aspect.

[0169] The readable storage medium may be more specifically adopted, including but not limited to: portable disk, hard disk, random access memory, read-only memory, erasable programmable read-only memory, optical storage device, magnetic storage device, or any suitable combination thereof.

[0170] In a possible implementation, the present invention can also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform steps implementing the AI ​​recommendation ordering method provided in the first aspect.

[0171] The program code for executing the present invention can be written in any combination of one or more programming languages. The program code can be executed entirely on the user device, partially on the user device, as a standalone software package, partially on the user device and partially on a remote device, or entirely on a remote device.

[0172] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0173] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. An AI-recommended order placement method, characterized in that, include: At least two user intents are extracted from real-time acquired natural language text. The priority weights of the user intents are determined based on the context and the user's historical behavior. The priority weights include an urgency weight to reflect the urgency of the user intent and a contextual association weight to characterize the degree to which the current user's historical dialogue content complements the user intent. The context refers to the current user's historical dialogue content. If the context information includes intent weight-related information, its corresponding contextual association weight is higher. Based on the priority weights, a target model is determined from the pre-acquired entity extraction model, and entity information corresponding to the user intent is extracted through the target model. Different priority weights correspond to different target models. Based on the entity information, a recommendation card for candidate products is sent to the user. The recommendation card includes a purchase interface for the candidate products so that the user can complete the purchase operation. The step of determining the target model from the pre-acquired entity extraction models based on the priority weights includes: In response to the priority weight being greater than or equal to the second weight threshold, the first model is selected as the target model; In response to the priority weight being less than the second weight threshold, the second model is selected as the target model, and the model size and computational cost of the second model are less than those of the first model; The first model is a high-precision model, and the second model is a lightweight model.

2. The AI-recommended order placement method according to claim 1, characterized in that, The process of determining the priority weight of the user intent based on context and the user's historical behavior includes: The urgency weight corresponding to the user intent is determined based on the context and the first keyword in each user intent; In response to any of the emergency weights being greater than or equal to the first weight threshold, the emergency weight is used as the priority weight.

3. The AI-recommended order placement method according to claim 2, characterized in that, The method of determining the priority weight of the user intent based on context and the user's historical behavior also includes: In response to each of the urgency weights being less than the first weight threshold, based on the context and the user's historical behavior, an explicit weight and a context-related weight corresponding to the user's intent are determined; The priority weight corresponding to the user intent is determined based on the explicit weight, contextual weight, and urgency weight.

4. The AI-recommended order placement method according to claim 3, characterized in that, The determination of explicit weights and context-related weights corresponding to the user's intent based on the context and the user's historical behavior includes: The explicit weight of the user intent is determined based on the second keyword and semantic structure in the user intent; The contextual association weight of the user's intent is determined based on the context and the user's historical behavior.

5. The AI-recommended order placement method according to claim 4, characterized in that, The method further includes: If the explicit weight of the user's intent is less than the third weight threshold, an intent confirmation question is provided to the user. In response to the answer to the user's input intent confirmation question, the user intent is re-determined based on the answer and the natural language text.

6. The AI-recommended order placement method according to claim 1, characterized in that, The step of sending a recommendation card for candidate products to the user based on the entity information includes: Acquire environmental data and initial user behavior data, and determine the user's first shopping personality based on the environmental data, the behavior data, and a pre-constructed feature-personality-product knowledge graph; Based on the entity information, a first recommendation strategy corresponding to the first shopping personality is obtained from the feature-personality-product knowledge graph, and product recommendations are made based on the first recommendation strategy. Obtain user behavior data based on recommended product feedback, and determine the user's second shopping personality based on the environmental data, the behavior data, and the feature-personality-product knowledge graph; Based on the entity information, a second recommendation strategy corresponding to the second shopping personality is obtained from the feature-personality-product knowledge graph, and products are recommended based on the second recommendation strategy.

7. The AI-recommended order placement method according to claim 1, characterized in that, The step of sending a recommendation card for candidate products to the user based on the entity information includes: The recommended card type is determined from the first mapping relationship based on the type of the candidate product, wherein the first mapping relationship is a pre-stored mapping relationship between the candidate product type and the recommended card type; Recommended card information is determined from the second mapping relationship based on the recommended card type. The second mapping relationship is a pre-stored mapping relationship between recommended card types and recommended card information.

8. An AI-recommended order placement system, characterized in that, include: Weight determination module: used to extract at least two user intents from real-time acquired natural language text, and determine the priority weight of the user intents based on context and the user's historical behavior. The priority weights include an urgency weight to reflect the urgency of the user intents, and a context association weight to characterize the degree to which the current user's historical dialogue content supplements the user intents. The context refers to the current user's historical dialogue content. If the context information includes intent weight-related information, its corresponding context association weight is higher. Entity extraction module: used to determine a target model from the pre-acquired entity extraction model based on the priority weights, and extract entity information corresponding to the user intent through the target model, wherein different priority weights correspond to different target models; Recommendation module: used to send recommendation cards of candidate products to users based on the entity information, the recommendation cards including purchase interfaces of the candidate products for users to complete the purchase operation; the step of determining the target model from the pre-acquired entity extraction model based on the priority weight includes: In response to the priority weight being greater than or equal to the second weight threshold, the first model is selected as the target model; In response to the priority weight being less than the second weight threshold, the second model is selected as the target model, and the model size and computational cost of the second model are less than those of the first model; The first model is a high-precision model, and the second model is a lightweight model.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor, when executing the computer program, implements the AI ​​recommendation order placement method as described in any one of claims 1 to 7.

Images