Resource exchange recommendation method and device, medium and product

Abstract

The invention discloses a resource exchange recommendation method and device, a medium and a product, which are applied to the field of financial science and technology, and comprise the following steps: receiving a natural language input by a target user, and analyzing the natural language to obtain a resource exchange intention; extracting multi-dimensional constraint features of the target user, and evaluating the resource exchange intention according to the multi-dimensional constraint features through an intention confidence evaluation model to obtain a confidence score; and when the resource exchange intention is determined to be a fuzzy intention according to the confidence score, performing resource exchange recommendation on the target user according to the multi-dimensional constraint features. The accuracy of intention recognition is improved by extracting the multi-dimensional constraint features and obtaining the confidence score through the intention confidence evaluation model. When the resource exchange intention is determined to be the fuzzy intention according to the confidence score, the portrait label is extracted and recommended, the real preference behind the fuzzy demand can be accurately mined, the recommendation fits the reality of the user, the invalid recommendation is reduced, and the exchange intention and experience of the user are improved.

Description

Technical Field

[0001] This invention relates to the field of financial technology, and in particular to a resource exchange recommendation method, device, medium, and product. Background Technology

[0002] With the rapid development of mobile financial services, mobile banking has become the core platform for users to conduct financial transactions and enjoy value-added services. Resource exchange, as an important value-added service to enhance user stickiness and revitalize user assets, directly affects user experience and platform activity through its service quality.

[0003] Currently, mobile banking resource redemption scenarios mainly rely on two implementation methods: one is a passive response mode based on keyword search, where users need to actively input specific information such as the name, category, or brand of the resource to be redeemed, and the system returns a list of resource redemption targets based on keyword matching; the other is a recommendation mechanism based on simple rules, such as displaying preset popular resource redemption targets according to points tiers, or randomly filtering and sorting resource redemption targets to display to users, or setting up a list of favorite items to display popular resource redemption targets.

[0004] Existing resource redemption technologies rely entirely on users providing explicit search keywords. When users express only vague needs, effective intent recognition and guidance are impossible. Furthermore, existing technologies only recommend resources based on fixed rules or popularity, resulting in a mismatch between recommended redemption targets and users' actual needs. In addition, existing technologies depend on manually preset rules, which cannot adapt to different users' personalized scenarios, differences in remaining resources, and changes in consumption habits. Moreover, traditional recommendation models require a large amount of historical behavioral data, and their effectiveness drops sharply in scenarios with new users or users who redeem resources infrequently, ultimately leading to a poor user experience and low resource redemption rates. Summary of the Invention

[0005] This invention provides a resource exchange recommendation method, device, medium, and product. By parsing users' natural language and combining multi-dimensional constraint features to assess the clarity of exchange intentions, and then implementing personalized recommendations with tags for ambiguous intentions, it solves the technical problems of inaccurate user intention recognition, low recommendation relevance, opaque rules, and low exchange efficiency in traditional resource exchange.

[0006] According to one aspect of the present invention, a resource exchange recommendation method is provided, the method comprising: Receive natural language input from the target user, parse the natural language, and obtain the resource exchange intention; Multidimensional constraint features of target users are extracted, and resource exchange intentions are evaluated based on these multidimensional constraint features using an intent confidence assessment model to obtain a confidence score. The multidimensional constraint features include intent salience features, user attribute features, user behavior features, and resource exchange constraint features. When the resource redemption intention is determined to be fuzzy based on the confidence score, resource redemption recommendations are made to the target user based on multidimensional constraint features.

[0007] Optionally, the natural language is parsed to obtain the resource exchange intent, including: generating a first prompt word, calling a large model to perform semantic parsing of the natural language based on the first prompt word, filtering out redundant information unrelated to resource exchange, extracting intent elements related to resource exchange, integrating the intent elements, and generating the resource exchange intent. The intent elements include behavioral verb elements, specific targeting elements, and constraint description elements.

[0008] The advantage of this setup is that it accurately filters relevant and effective information on resource exchange, eliminates redundant interference, and ensures the relevance and completeness of the extracted intent elements.

[0009] Optionally, the intention to exchange resources is evaluated based on multi-dimensional constraint features using an intention confidence assessment model to obtain a confidence score. This includes: normalizing the multi-dimensional constraint features and each intention element in the resource exchange intention to generate standard input features; generating a second prompt word; calling the large model based on the second prompt word and using a dynamic weight adjustment mechanism to determine the feature weights corresponding to each standard input feature based on preset association rules, wherein the second prompt word includes preset rules; and inputting each standard input feature and its corresponding feature weights into a pre-trained intention confidence assessment model to calculate the confidence score.

[0010] The advantages of this setup are as follows: Normalization standardizes input features, ensuring consistency in model input. Dynamic weight adjustment enhances the relevance of the evaluation. The calculated confidence score reflects the clarity of resource exchange intentions, providing a clear criterion for subsequent differentiated processing.

[0011] Optionally, resource redemption recommendations are made for target users based on multi-dimensional constraint features, including: generating third prompt words; calling a large model based on the third prompt words to generate profile tags according to multi-dimensional constraint features; obtaining a resource redemption target pool, wherein the resource redemption target pool includes each resource redemption target and its corresponding resource redemption target attributes, required resources, inventory status, and resource redemption target tags; based on the pre-established association mapping between profile tags and resource redemption target tags, filtering resource redemption targets in the resource redemption target pool that match the target user, and generating a candidate target set; after filtering the candidate target set based on resource redemption constraint features, generating resource recommendation results.

[0012] The advantages of this setup are: the generation of user profile tags achieves the structuring and visualization of user attribute characteristics, facilitating subsequent matching; and tag association mapping enables precise matching of resource exchange targets with user needs, improving recommendation relevance.

[0013] Optionally, after filtering the candidate target set based on resource exchange constraint features, a resource recommendation result is generated, including: determining exchange constraint conditions through resource exchange constraint features, filtering out resource exchange targets in the candidate target set whose required resources exceed the exchange constraint conditions, generating a subset of candidate targets; and generating a resource recommendation result based on the subset of candidate targets.

[0014] The advantage of this setting is that by determining the exchange constraints through resource exchange constraint features, the constraints are made more in line with users' points usage habits and actual capabilities, effectively eliminating invalid resource exchange targets that users cannot exchange, and generating a precise subset of candidate targets.

[0015] Optionally, generating resource recommendation results based on a subset of candidate targets includes: determining the historical redemption count of each resource redemption target in the subset of candidate targets; sorting the subset of candidate targets in descending order of historical redemption count to generate a sorted subset of candidate targets; selecting a specified number of resource redemption targets from the subset of candidate targets as final recommended targets; generating a fourth prompt word; and using the fourth prompt word to call the large model to generate interpretable text based on multidimensional constraint features and the final recommended targets; and combining the interpretable text and the final recommended targets to generate resource recommendation results.

[0016] The advantages of this setup are as follows: The historical redemption frequency ranking provides an objective basis for the popularity of recommended items, improving the rationality of the recommendations. The generated interpretable wording makes the recommendation logic transparent, solving the problem of unclear traditional recommendation rules. Combining the wording with the items makes the recommendation results more personalized and persuasive, increasing users' willingness to redeem.

[0017] Optionally, the method also includes: when the resource redemption intention is determined to be a clear intention based on the confidence score, calling the query interface to query the target in the resource redemption target pool that matches the resource redemption intention, and using the target as the resource recommendation result.

[0018] The advantages of this setup are: the query interface call enables rapid matching of target items, improves recommendation efficiency, avoids invalid recommendations, and increases the resource exchange conversion rate in scenarios with clear intent.

[0019] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform a resource exchange recommendation method according to any embodiment of the present invention.

[0020] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement a resource exchange recommendation method according to any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer program product is provided, the computer program product comprising a computer program that, when executed by a processor, implements a resource exchange recommendation method according to any embodiment of the present invention.

[0022] The technical solution of this invention, by receiving the target user's natural language and parsing their resource redemption intent, can directly capture the user's core needs, filter redundant information, and improve the convenience of interaction by eliminating the need for users to input in a fixed format. By extracting multi-dimensional constraint features and obtaining a score through an intent confidence evaluation model, the accuracy of intent recognition is improved. When a vague intent is determined based on the confidence score, profile tags are extracted and recommendations are made, which can accurately uncover the true preferences behind vague needs, ensuring that recommendations match the user's actual situation, reducing invalid recommendations, and improving the user's willingness to redeem and overall experience.

[0023] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart of a resource exchange recommendation method provided in Embodiment 1 of the present invention; Figure 2 This is a flowchart of another resource exchange recommendation method provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of a resource exchange recommendation device according to Embodiment 3 of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device that implements a resource exchange recommendation method according to an embodiment of the present invention. Detailed Implementation

[0026] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0027] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0028] Example 1 Figure 1 This is a flowchart of a resource exchange recommendation method provided in Embodiment 1 of the present invention. This embodiment is applicable to points-based recommendation scenarios. The method can be executed by a resource exchange recommendation device, which can be implemented in hardware and / or software and can be configured within an intelligent agent. Figure 1 As shown, the method includes: S110: Receive natural language input from the target user, parse the natural language, and obtain the resource exchange intention.

[0029] Natural language refers to the text content entered by the target user in a daily communication manner, such as "My points are about to expire, I want to exchange them for something useful" or "I want to use my points to exchange for a mobile phone." Resource exchange intent refers to the target user's desire to use resources to exchange for an item, conveyed through natural language input, and is divided into two categories: explicit intent and vague intent.

[0030] Specifically, an intelligent agent refers to an intelligent agent system built on a large language model, possessing natural language understanding, logical reasoning, and decision-making capabilities. It can simulate human dialogue and perform complex tasks. The large-model intelligent agent can receive text content input by users in everyday conversational manner and perform real-time semantic parsing of natural language through a dialogue engine to extract basic intent elements. These elements include, for example, the mentioned brand "Brand A," the name of the resource exchange target "mobile phone," the category "electronic product," the action verb "exchange," and the budget description "useful items." By identifying these intent elements, the user's core resource exchange needs—that is, their resource exchange intent—can be clearly understood.

[0031] Optionally, the natural language is parsed to obtain the resource exchange intent, including: generating a first prompt word, calling a large model to perform semantic parsing of the natural language based on the first prompt word, filtering out redundant information unrelated to resource exchange, extracting intent elements related to resource exchange, integrating the intent elements, and generating the resource exchange intent. The intent elements include behavioral verb elements, specific targeting elements, and constraint description elements.

[0032] Specifically, firstly, the intelligent system generates a first prompt word based on the semantic parsing requirements of the resource exchange scenario. The design principle of the first prompt word is based on the large model's accurate recognition capability of semantics in a specific domain, clearly guiding the large model to focus on the core task of resource exchange and avoiding the generalization problem of general semantic parsing. The first prompt word must cover filtering redundant information, extracting three types of core intent elements, and structurally integrating core instructions, ensuring that the large model performs parsing operations according to a unified standard. During the parsing process, the large model first filters out redundant information unrelated to resource exchange, such as irrelevant chatter or descriptive statements that may be included in the user input, retaining only core information fragments related to resource exchange. Then, it extracts intent elements related to resource exchange from the retained valid information. Intent elements include behavioral verb elements, specific targeting elements, and constraint description elements. Among them, behavioral verb elements refer to words used by the user to express actions related to resource exchange, such as "exchange," "redeem," and "exchange for," clearly defining the user's core behavioral needs. Specific targeting elements are specific information mentioned by the user related to the exchange target, such as brand, name of the exchange target, and category, directly pointing to the specific exchange target. The constraint description elements are the user's limiting expressions of their redemption needs, such as "practical," "points are about to expire," and "high cost-performance ratio," reflecting the user's additional needs or scenario constraints. After extraction, the large model integrates the various intent elements, combining them with the contextual logic of semantic parsing to identify the user's core needs, ultimately generating a clear and complete resource redemption intent. This intent includes both the user's core behavior and possible specific goals or constraints, providing an accurate basis for subsequent intent confidence assessment. For example, if a user inputs "points are about to expire, I want to redeem a practical electronic product from brand A," the parsing will filter out redundant information and extract the behavioral verb "redeem," the specific target elements "brand A" and "electronic product," and the constraint description elements "points are about to expire" and "practical." The resulting resource redemption intent is: The user wants to redeem a practical electronic product from brand A because their points are about to expire.

[0033] S120. Extract the multidimensional constraint features of the target user, and evaluate the resource exchange intention based on the multidimensional constraint features using the intent confidence assessment model to obtain a confidence score.

[0034] The multidimensional constraint features include intent salience features, user attribute features, user behavior features, and resource exchange constraint features. Multidimensional constraint features are a multi-dimensional data set used to construct user profiles and support intent assessment and recommendation inference. Intent salience features refer to the clarity and prominence of resource exchange-related intent elements parsed from the user's natural language input, reflecting the clarity of the user's exchange request, including the completeness, clarity, and weight of intent elements in the input text. User attribute features refer to the static attribute information of the target user, such as age, gender, membership level, and occupation; these are data that characterize the user's basic features. User behavior features refer to the feature set that records the target user's past dynamic behavior trajectory, such as consumption-related information extracted from the target user's transaction data, including the proportion of consumption amount, frequency of consumption, and recent consumption time in scenarios such as supermarkets, restaurants, and transportation over the past N months (e.g., 3 / 6 months). Resource exchange constraint features refer to the key feature set that limits the user's resource exchange ability and scope, such as dynamic data related to user points and exchange behavior, including current available points balance, historical exchange records, and exchange behavior patterns. The intent confidence assessment model is a dynamic model used to quantitatively assess the clarity of a user's intent to exchange resources.

[0035] Optionally, the intention to exchange resources is evaluated based on multi-dimensional constraint features using an intention confidence assessment model to obtain a confidence score. This includes: normalizing the multi-dimensional constraint features and each intention element in the resource exchange intention to generate standard input features; generating a second prompt word; calling the large model based on the second prompt word and using a dynamic weight adjustment mechanism to determine the feature weights corresponding to each standard input feature based on preset association rules, wherein the second prompt word includes preset rules; and inputting each standard input feature and its corresponding feature weights into a pre-trained intention confidence assessment model to calculate the confidence score.

[0036] It is known that multidimensional constraint features and intent elements differ in type and magnitude. For example, user age is numerical data, while occupation is categorical data. Furthermore, the numerical ranges of consumption amount percentage and points balance are not uniform. These data differences can interfere with the model's judgment of feature importance, leading to biased evaluation results. Therefore, normalization is needed to eliminate the influence of data dimensions and make different types of features comparable. Multidimensional constraint features include user attribute features, user behavior features, and resource redemption constraint features. The intent elements in resource redemption intent include behavioral verb elements, specific targeting elements, and constraint description elements. During processing, numerical data, such as consumption amount percentage, points balance, and age, are mapped to a uniform value range, such as 0-1. Categorical data, such as gender, occupation, and behavioral verbs, undergo one-hot encoding transformation, ultimately generating standard input features with a uniform format that can be directly input into the model.

[0037] It's important to note that the correlation between different features and the clarity of a user's resource redemption intention varies depending on the user's input scenario. Fixed weights cannot accurately capture the user's true needs; therefore, dynamic adjustments based on preset rules are necessary to give key features higher weights and improve evaluation accuracy. Preset association rules are logical rules pre-defined based on the characteristics of the resource redemption scenario. For example, when user input contains ambiguous words, user behavior features and historical redemption preferences play a more significant role in determining the user's true intention, thus requiring increased weights for such features. Specifically, the AI ​​generates a second prompt word and calls the large model to determine feature weights. The core design of the second prompt word is to transform the preset association rules into reasoning instructions understandable by the large model. The preset association rules are based on the technical disclosure's analysis of the correlation between user behavior and redemption intention, such as prioritizing resource redemption constraint features when resources are nearing expiration. After the large model is invoked, it first identifies the key information of the user's input intention elements and multi-dimensional constraint features, and then matches them with the corresponding preset association rules. For example, if the system detects that the user input contains vague terms like "practical," and the query shows that the user's supermarket spending over the past three months accounts for more than 60% of their total spending, then the weight of the standard input features related to supermarket spending and historical redemption of daily necessities will be increased by 50% according to the rules. If the user input explicitly mentions a specific brand or resource redemption item, the weight of the standard input features corresponding to the specific element will be significantly increased, while the weight of other less relevant features will be reduced accordingly.

[0038] The pre-trained intent confidence assessment model can be a tree model, capable of learning the mapping relationship between features and intent clarity. Combined with dynamically adjusted feature weights, it can accurately quantify the clarity of user intent. The intent confidence assessment model is pre-trained on a large amount of historical data, mastering the correspondence between different feature combinations and confidence scores. After inputting standard input features and feature weights, the model performs weighted calculations according to the preset algorithm logic, based on the weight and specific value of each feature. For example, features with high weights that are clear and complete will significantly increase the score, while vague constraint descriptions will decrease the score. The final output is a confidence score ranging from 0 to 100. A higher score indicates a clearer user intent to redeem resources, providing a quantitative basis for subsequently distinguishing between clear and vague intents and triggering different processing flows.

[0039] S130. When the resource exchange intention is determined to be fuzzy based on the confidence score, resource exchange recommendations are made to the target user based on the multidimensional constraint features.

[0040] The confidence score, calculated by the intent confidence assessment model, is a quantitative indicator reflecting the clarity of a user's resource redemption intent. It ranges from 0 to 100, with higher scores indicating clearer user goals. This score triggers different subsequent processing steps. The clarity of a resource redemption intent can be determined by a preset score, a baseline value (e.g., 75) set to distinguish between clear and vague intents. When the confidence score is lower than the preset score, it is considered a vague intent, triggering the recommendation logic. A vague intent refers to a resource redemption intent with a confidence score below a preset threshold. This manifests as user input lacking key elements such as a clear brand, the name of the resource redemption target, or a specific category, containing only vague descriptions like "useful" or "good." Multi-dimensional data reasoning is needed to uncover the user's true needs. User profile tags are labels extracted based on multi-dimensional constrained features that characterize a user's core needs, such as family life users, frequent travelers, and tech enthusiasts. Resource redemption recommendations are personalized service outputs for users with vague intents. Based on user profile tags and points balance constraints, and through scenario-based reasoning, a list of resource redemption targets with recommended reasons is ultimately output.

[0041] Optionally, the method also includes: when the resource redemption intention is determined to be a clear intention based on the confidence score, calling the query interface to query the target in the resource redemption target pool that matches the resource redemption intention, and using the target as the resource recommendation result.

[0042] When the confidence score is greater than or equal to the preset score, it means that the clarity of the user's intention to exchange resources has reached the system's judgment standard. At this time, there is no need to start a complex recommendation reasoning process. Instead, the system can quickly match user needs through precise query. That is, it can use clear intent elements as query conditions to efficiently filter target objects in the resource exchange target pool, taking into account both response speed and result accuracy.

[0043] Specifically, when the confidence score calculated by the intent confidence assessment model is greater than or equal to the preset score, the user's resource redemption intent is determined to be explicit intent. Explicit intent typically possesses complete and clear intent elements. For example, if a user enters "I want to use points to redeem a P-series phone from brand A," it includes the explicit brand "Brand A," the redemption target series "P-series," and the category "phone." The analytical element strength is high, and after model evaluation, the confidence score reaches the preset standard, thus being classified as explicit intent.

[0044] The query interface refers to the high-concurrency, low-latency resource redemption query interface in the service support layer. This interface supports queries based on multiple conditions, including brand, category, resource redemption item name, and points range, meeting users' needs for query speed. After determining the specific intent, the system extracts key intent elements from the resource redemption intent, such as brand, resource redemption item name, and category. These intent elements are then converted into query parameters recognizable by the interface. For example, "Brand A," "Mobile Phone," and "P Series" might be used as core query conditions. Simultaneously, a points range constraint is set based on the user's current available points balance. The system then calls the resource redemption query interface to initiate a query request to the resource redemption item pool. The resource redemption item pool contains data on all available resource redemption items, covering attributes, required resources, and inventory status. After receiving the query parameters, the query interface filters through the resource redemption item pool, matching resource redemption items that simultaneously meet all query conditions, including brand, category, and points range—the target items that match the resource redemption intent. Finally, the system will compile the target items into a list of resource exchange targets. The list includes information such as the image of the resource exchange target, the required resources, and the details of the resource exchange target. This list will be returned to the user as a resource recommendation result, allowing the user to directly select the resource exchange target that meets their specific needs for exchange.

[0045] The technical solution of this invention, by receiving the target user's natural language and parsing their resource redemption intent, can directly capture the user's core needs, filter redundant information, and improve the convenience of interaction by eliminating the need for users to input in a fixed format. By extracting multi-dimensional constraint features and obtaining a score through an intent confidence evaluation model, the accuracy of intent recognition is improved. When a vague intent is determined based on the confidence score, profile tags are extracted and recommendations are made, which can accurately uncover the true preferences behind vague needs, ensuring that recommendations match the user's actual situation, reducing invalid recommendations, and improving the user's willingness to redeem and overall experience.

[0046] Example 2 Figure 2 This is a flowchart of a resource exchange recommendation method provided in Embodiment 2 of the present invention. This embodiment adds a specific process to Embodiment 1, which involves extracting profile tags based on multi-dimensional constraint features and recommending resource exchanges to target users based on these profile tags. The specific content of steps S250-S260 is largely the same as steps S120-S130 in Embodiment 1, and therefore will not be repeated in this embodiment. Figure 2 As shown, the method includes: S210. Receive natural language input from the target user, parse the natural language, and obtain the resource exchange intention.

[0047] Optionally, the natural language is parsed to obtain the resource exchange intent, including: generating a first prompt word, calling a large model to perform semantic parsing of the natural language based on the first prompt word, filtering out redundant information unrelated to resource exchange, extracting intent elements related to resource exchange, integrating the intent elements, and generating the resource exchange intent. The intent elements include behavioral verb elements, specific targeting elements, and constraint description elements.

[0048] S220. Extract the multidimensional constraint features of the target user, and evaluate the resource exchange intention based on the multidimensional constraint features using the intent confidence assessment model to obtain a confidence score.

[0049] Among them, the multidimensional constraint features include intent salience features, user attribute features, user behavior features, and resource exchange constraint features.

[0050] Optionally, the intention to exchange resources is evaluated based on multi-dimensional constraint features using an intention confidence assessment model to obtain a confidence score. This includes: normalizing the multi-dimensional constraint features and each intention element in the resource exchange intention to generate standard input features; generating a second prompt word; calling the large model based on the second prompt word and using a dynamic weight adjustment mechanism to determine the feature weights corresponding to each standard input feature based on preset association rules, wherein the second prompt word includes preset rules; and inputting each standard input feature and its corresponding feature weights into a pre-trained intention confidence assessment model to calculate the confidence score.

[0051] S230. When the resource exchange intention is determined to be ambiguous based on the confidence score, a third prompt word is generated. Based on the third prompt word, the large model is called to generate profile labels according to multi-dimensional constraint features.

[0052] Optionally, the method also includes: when the resource redemption intention is determined to be a clear intention based on the confidence score, calling the query interface to query the target in the resource redemption target pool that matches the resource redemption intention, and using the target as the resource recommendation result.

[0053] The multi-dimensional constraint features include static user attribute features, dynamic user behavior features, and resource exchange constraint features. First, the intelligent system generates a third prompt word. The design principle of this third prompt word is to anchor the specific needs of the bank's resource exchange scenario, clearly guiding the large model to focus on the core data dimensions of the multi-dimensional constraint features. The prompt word implicitly contains the thought chain reasoning requirements for consumer characteristic profiling, ensuring that the large model does not deviate from the scenario-based analysis direction. Then, the large model uses feature aggregation and inductive logic to comprehensively analyze the multi-dimensional constraint features. For example, if the target user's supermarket spending accounts for more than 60% of their spending in the past 3 months and they have previously redeemed kitchenware, they will be categorized as a "family life user"; if the target user frequently spends on travel and repeatedly redeems portable products, a "high-frequency travel user" tag will be generated; if the target user has redeemed electronic products 3 times in the past 6 months, they will be labeled as a "technology product enthusiast."

[0054] S240. Obtain the resource exchange target pool, wherein the resource exchange target pool includes each resource exchange target and its corresponding resource exchange target attributes, required resources, inventory status and resource exchange target tags.

[0055] The resource redemption pool is a collection of all redeemable resources in the bank's resource redemption service. It includes not only basic information about each resource redemption item, but also key screening and matching dimensions. These include the attributes of the resource redemption item itself, such as brand, category, specifications, and features; the number of points required to redeem the resource redemption item; the current inventory status, such as availability and remaining stock, to avoid recommending out-of-stock resources redemption items that could negatively impact user experience; and preset resource redemption item tags for each resource redemption item, such as kitchen appliances and travel gear, which are set by operations personnel based on the characteristics of the resource redemption item and the target customer group.

[0056] S250. Based on the pre-established association mapping between profile tags and resource exchange target tags, filter resource exchange targets in the resource exchange target pool that match the target user and generate a candidate target set.

[0057] Specifically, association mapping refers to the corresponding logic based on the matching of user needs and resource redemption targets. It achieves accurate initial screening through tag matching, reducing the number of resource redemption targets in subsequent screenings and improving recommendation efficiency. The system first builds a tag association rule library, clarifying the tags of suitable resource redemption targets corresponding to different user profile tags. For example, the "family life user" tag is associated with tags for resource redemption targets such as kitchen appliances and daily necessities; the "frequent travel user" tag is associated with tags for resource redemption targets such as "portable devices" and "travel gear"; and the "technology product enthusiast" tag is associated with tags for resource redemption targets such as "electronic products" and "smart devices". Finally, the generated user profile tags are compared one by one with the resource redemption target tags in the resource redemption target pool, filtering out all resource redemption targets whose tags match the user profile tags, forming a candidate target set.

[0058] S260. Determine the exchange constraint conditions through resource exchange constraint features, filter out resource exchange targets in the candidate target set whose required resources exceed the exchange constraint conditions, and generate a subset of candidate targets.

[0059] The resource exchange constraint features include data such as the user's current available points balance, points consumption values ​​in historical exchange records, and exchange behavior patterns. When determining exchange constraints, the user's current available points balance is used as the basic constraint threshold. Available points directly determine the upper limit of resource exchange targets that a user can actually exchange, making it a core hard indicator for the resource exchange scenario. Simultaneously, the system will flexibly adjust based on the user's points consumption patterns. For example, if a user exhibits a behavior characteristic of points nearing expiration, the exchange constraints will be appropriately broadened, prioritizing resource exchange targets whose points consumption values ​​are close to the user's available points, allowing the user to fully utilize the soon-to-expire points. If the user's historical exchange records show a fixed preference for points consumption ranges, this will also be considered in the constraint settings to ensure the constraints better align with the user's exchange habits. Then, the system will verify the required resources for each resource exchange target in the candidate target set, filtering out all resource exchange targets whose required resources exceed the exchange constraint threshold, retaining only those resource exchange targets whose required resources are less than or equal to the exchange constraint threshold, forming a subset of candidate targets.

[0060] S270. Determine the historical exchange count of each resource exchange target in the candidate target subset, sort the candidate target subset in descending order of historical exchange count, and generate the sorted candidate target subset.

[0061] Among these metrics, the historical redemption count directly reflects the popularity and practicality of resource redemption targets, serving as a concentrated representation of user preferences. Prioritizing resource redemption targets with high historical redemption counts improves the acceptance of recommendation results. The system can retrieve historical redemption count data for each resource redemption target from the resource redemption target pool through the real-time data interface of the service support layer. This historical redemption count data includes statistics on all user redemption records for that resource redemption target within a certain period, such as the past three months. Then, the resource redemption targets in the candidate target subset are sorted according to their historical redemption count from highest to lowest, with resource redemption targets ranking higher based on their historical redemption count. This results in a sorted subset of candidate targets, ensuring that resource redemption targets with high market recognition are prioritized.

[0062] S280. Select a specified number of resource exchange targets from a subset of candidate targets as the final recommended targets.

[0063] It is known that selecting a specified number of resource exchange targets from a subset of the sorted candidate targets as the final recommended targets can avoid overwhelming users with too many recommended resource exchange targets, while focusing on core, high-quality resource exchange targets and improving recommendation efficiency. The specified number is preset based on user experience research and business scenario requirements; for example, it can be set to 3, which ensures both the richness of recommendations and avoids overwhelming users with choices. The system will select the specified number of resource exchange targets sequentially from the sorted results as the final recommended targets. The final recommended targets not only meet the user's redemption constraints and need matching, but also have high market popularity.

[0064] S290. Generate the fourth prompt word, and based on the fourth prompt word, call the large model to generate interpretable text based on multi-dimensional constraint features and the final recommendation target.

[0065] Specifically, the resource exchange recommendation process utilizes the step-by-step reasoning logic of the thought chain technology to establish a clear connection between user attribute characteristics and the advantages of the resource exchange target. This allows users to understand why the resource exchange target is recommended, eliminating the mystery surrounding the recommendation and enhancing trust. First, the intelligent system generates a fourth prompt word. This fourth prompt word must clearly guide the large model to establish a link between user characteristics, recommendation logic, and the value of the target, while anchoring to the service attributes of the bank's resource exchange scenario. The fourth prompt word implicitly contains three core instructions: first, it must be linked to key data from the multi-dimensional constraint features to avoid generalized recommendation reasons; second, it must combine the core attributes of the final recommended target to explain the fit between the recommended target and the user's needs; and third, the language style must be concise, easy to understand, and relevant to the bank's service scenario, avoiding technical jargon to ensure users quickly understand the recommendation logic. For example, the fourth prompt word could be: "Based on the user's high frequency of offline dining consumption and sufficient resource balance, combined with the dining scenario adaptation function of the recommended target, a natural and easy-to-understand recommendation reason is generated." The large model deeply integrates multi-dimensional constraint features, including user attribute features, user behavior features, and resource redemption constraint features. It then combines these with the attributes of the final recommended item, generating the corresponding dialogue based on the inference chain of user attribute features, demand matching, and the value of the resource redemption item. For example, if a user's dining consumption has significantly increased in the past three months, and the final recommended item is a capsule coffee machine, the interpretable text would be generated as follows: "We have recognized a significant increase in your dining consumption in the past three months, and you have sufficient points. We recommend this capsule coffee machine so you can enjoy coffee at home." If the user is a frequent traveler and the recommended resource redemption item is a portable power bank, the interpretable text would be: "Based on your frequent travel characteristics and past preferences for redeeming electronic products, we recommend this portable power bank."

[0066] S300: Combine the interpretable text with the final recommendation target to generate resource recommendation results.

[0067] Specifically, combining explanatory text with the final recommended resource involves integrating the basic information of the resource redemption target with personalized recommendation reasons to form a complete recommendation presentation. This allows users to obtain key information about the resource redemption target while clearly understanding the recommendation logic, thus increasing the conversion rate. The final resource recommendation results can be presented in the form of resource redemption target cards with reasons. Each card includes basic information such as an image of the resource redemption target, the required resource, and details of the resource redemption target, along with generated explanatory text. Furthermore, the resource redemption target cards will be displayed according to the previous sorting results, ensuring that high-popularity and highly relevant resources are prioritized, allowing users to intuitively see the core value of the recommended resource redemption target and its relevance to themselves, enabling them to quickly make a redemption decision.

[0068] The technical solution of this invention generates user profile tags tailored to specific needs by profiling multi-dimensional constraint features, accurately extracting user consumption preferences and redemption tendencies, thus laying a core foundation for personalized recommendations. By establishing tag association mappings to generate a candidate set, it achieves precise matching between user needs and resource redemption targets, significantly improving recommendation relevance. Redemption constraints are determined and resource redemption targets are filtered through resource redemption constraint features, ensuring that recommended resource redemption targets match the user's points accumulation ability and avoiding invalid recommendations. Resources redemption targets are sorted by historical redemption frequency and a specified number are selected, prioritizing those with high market recognition and strong user acceptance. By combining interpretable text with resource redemption targets to generate recommendation results, users can intuitively perceive the suitability value of resource redemption targets, enhancing their redemption willingness and experience.

[0069] Example 3 Figure 3 This is a schematic diagram of a resource exchange recommendation device provided in Embodiment 3 of the present invention. Figure 3 As shown, the device includes: a resource exchange intent generation module 310, used to receive natural language input from a target user, parse the natural language, and obtain a resource exchange intent; The confidence score calculation module 320 is used to extract the multidimensional constraint features of the target user, and evaluate the resource exchange intention based on the multidimensional constraint features through the intent confidence assessment model to obtain the confidence score. The multidimensional constraint features include intent salience features, user attribute features, user behavior features and resource exchange constraint features. The fuzzy intent redemption module 330 is used to recommend resources to the target user based on multi-dimensional constraint features when the resource redemption intent is determined to be fuzzy intent based on the confidence score.

[0070] Optionally, the resource exchange intent generation module 310 is specifically used to: generate a first prompt word, call the large model based on the first prompt word to perform semantic parsing of natural language, filter redundant information unrelated to resource exchange, extract intent elements related to resource exchange, integrate the intent elements, and generate a resource exchange intent. The intent elements include behavioral verb elements, specific targeting elements, and constraint description elements.

[0071] Optionally, the confidence score calculation module 320 is specifically used for: normalizing the multi-dimensional constraint features and the intent elements in the resource exchange intent to generate standard input features; generating second prompt words; calling the large model based on the second prompt words and using a dynamic weight adjustment mechanism to determine the feature weights corresponding to each standard input feature based on preset association rules, wherein the second prompt words include preset rules; inputting each standard input feature and its corresponding feature weights into the pre-trained intent confidence evaluation model to calculate the confidence score.

[0072] Optionally, the fuzzy intent redemption module 330 specifically includes: a profile tag generation unit, used to: generate a third prompt word, and based on the third prompt word, call the large model to generate profile tags according to multi-dimensional constraint features; a resource redemption target pool acquisition unit, used to: acquire a resource redemption target pool, wherein the resource redemption target pool includes each resource redemption target and its corresponding resource redemption target attributes, required resources, inventory status, and resource redemption target tags; a candidate target set generation unit, used to: based on the pre-established association mapping between profile tags and resource redemption target tags, filter resource redemption targets in the resource redemption target pool that match the target user, and generate a candidate target set; and a redemption constraint condition filtering unit, used to: after filtering the candidate target set based on resource redemption constraint features, generate resource recommendation results.

[0073] Optionally, the exchange constraint filtering unit is used to: determine exchange constraints through resource exchange constraint features, filter out resource exchange targets in the candidate target set whose required resources exceed the exchange constraints, generate a subset of candidate targets, and generate resource recommendation results based on the subset of candidate targets.

[0074] Optionally, the exchange constraint filtering unit is specifically used for: determining the historical exchange count of each resource exchange target in the candidate target subset; sorting the candidate target subset in descending order of historical exchange count to generate a sorted candidate target subset; selecting a specified number of resource exchange targets from the candidate target subset as the final recommended targets; generating a fourth prompt word; and based on the fourth prompt word, calling the large model to generate interpretable text based on multidimensional constraint features and the final recommended targets; and combining the interpretable text and the final recommended targets to generate resource recommendation results.

[0075] Optionally, the device further includes: a definite intent redemption module, used to call a query interface to query the resource redemption target pool for a target that matches the resource redemption intent when the resource redemption intent is determined to be a definite intent based on the confidence score, and to use the target as the resource recommendation result.

[0076] The technical solution of this invention, by receiving the target user's natural language and parsing their resource redemption intent, can directly capture the user's core needs, filter redundant information, and improve the convenience of interaction by eliminating the need for users to input in a fixed format. By extracting multi-dimensional constraint features and obtaining a score through an intent confidence evaluation model, the accuracy of intent recognition is improved. When a vague intent is determined based on the confidence score, profile tags are extracted and recommendations are made, which can accurately uncover the true preferences behind vague needs, ensuring that recommendations match the user's actual situation, reducing invalid recommendations, and improving the user's willingness to redeem and overall experience.

[0077] The resource exchange recommendation device provided in this embodiment of the invention can execute a resource exchange recommendation method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution.

[0078] Example 4 Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0079] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded into the RAM 13 from storage unit 18. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0080] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0081] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as a resource exchange recommendation method.

[0082] In some embodiments, a resource exchange recommendation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the resource exchange recommendation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform a resource exchange recommendation method by any other suitable means (e.g., by means of firmware).

[0083] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0084] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0085] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0086] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0087] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0088] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0089] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0090] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A resource exchange recommendation method, characterized in that, include: Receive natural language input from the target user, parse the natural language, and obtain the resource exchange intention; Extract multidimensional constraint features of the target user, and evaluate the resource exchange intention based on the multidimensional constraint features using an intent confidence assessment model to obtain a confidence score. The multidimensional constraint features include intent salience features, user attribute features, user behavior features, and resource exchange constraint features. When the resource exchange intention is determined to be fuzzy based on the confidence score, resource exchange recommendations are made to the target user based on the multidimensional constraint features.

2. The method according to claim 1, characterized in that, The process of parsing the natural language to obtain the resource exchange intent includes: A first prompt word is generated. Based on the first prompt word, a large model is invoked to perform semantic parsing on the natural language, filtering out redundant information unrelated to resource exchange, and extracting intent elements related to resource exchange. The intent elements are then integrated to generate a resource exchange intent. The intent elements include behavioral verb elements, specific targeting elements, and constraint description elements.

3. The method according to claim 2, characterized in that, The assessment of resource exchange intentions using the intent confidence evaluation model based on the multidimensional constraint features yields a confidence score, including: The multidimensional constraint features and resource exchange intentions are normalized to generate standard input features. A second prompt word is generated. Based on the second prompt word, the large model is called to determine the feature weights corresponding to each standard input feature through a dynamic weight adjustment mechanism based on preset association rules. The second prompt word includes preset rules. Each standard input feature and its corresponding feature weight are input into the pre-trained intent confidence assessment model to calculate the confidence score.

4. The method according to claim 1, characterized in that, The step of recommending resource redemption to the target user based on the multidimensional constraint features includes: Generate a third prompt word, and based on the third prompt word, call the large model to generate a profile label according to the multi-dimensional constraint features; Obtain a pool of resource exchange targets, wherein the pool of resource exchange targets includes each resource exchange target and its corresponding attributes, required resources, inventory status, and tags. Based on the pre-established association mapping between profile tags and resource exchange target tags, resource exchange targets that match the target user in the resource exchange target pool are filtered to generate a candidate target set; After filtering the candidate target set based on resource exchange constraint features, resource recommendation results are generated.

5. The method according to claim 4, characterized in that, After filtering the candidate target set based on resource exchange constraint features, the resource recommendation results are generated, including: The exchange constraints are determined by the resource exchange constraint characteristics. Resource exchange targets whose required resources exceed the exchange constraints are filtered out from the candidate target set, and a subset of candidate targets is generated. Resource recommendation results are generated based on the subset of candidate targets.

6. The method according to claim 5, characterized in that, The generation of resource recommendation results based on the subset of candidate targets includes: Determine the historical redemption count of each resource redemption target in the candidate target subset, and sort the candidate target subset in descending order of the historical redemption count to generate a sorted candidate target subset; Select a specified number of resource exchange targets from the subset of candidate targets as the final recommended targets; Generate a fourth prompt word, and based on the fourth prompt word, call the large model to generate interpretable text based on multi-dimensional constraint features and the final recommendation target; The interpretable text and the final recommendation target are combined to generate resource recommendation results.

7. The method according to claim 1, characterized in that, The method further includes: When the resource exchange intention is determined to be a clear intention based on the confidence score, the query interface is called to query the resource exchange target pool for target targets that match the resource exchange intention, and the target targets are used as resource recommendation results.

8. An electronic device, characterized in that, The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

9. A computer storage medium, characterized in that, The computer storage medium stores computer instructions that are used to cause a processor to execute the method of any one of claims 1-7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-7.

Images