Recommendation information generation method and device, equipment, storage medium and program product
By obtaining the session characteristics of the target object and the initial candidate service information set, and using preset recall rules and evaluation modules to generate the target service combination and its recommendation information, the problem of low accuracy and poor flexibility in the generation of recommended information in the prior art is solved, and more accurate and flexible recommendation information generation is achieved.
Patent Information
- Application Number
- CN202510168936.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-02-14
- Publication Date
- 2025-06-06
Smart Images
Figure CN120104868A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of data processing technology, and in particular to a method, device, equipment, storage medium and program product for generating recommendation information. Background Art
[0002] At present, the generation of recommendation information is mostly based on a solution-by-solution approach for users. During this process, conversational agents (such as intelligent customer service) will recommend some standard process solutions. However, the solution-by-solution process rarely considers the mutual influence between service information, and the processing process is relatively fixed, resulting in the generated recommendation information being difficult to accurately and flexibly solve user demands. Summary of the invention
[0003] In view of this, the present disclosure provides a method, apparatus, device, storage medium and program product for generating recommendation information to solve the problem of low accuracy and poor flexibility in generating recommendation information.
[0004] In a first aspect, the present disclosure provides a method for generating recommendation information, including: obtaining session characteristics of a target object and an initial candidate service information set; using a preset recall rule to recall a candidate service combination set that meets the session characteristics from the initial candidate service information set; evaluating each candidate service information in the candidate service combination set to determine the optimal service parameters of each candidate service information; determining a target service combination from the candidate service combination set according to the optimal service parameters and session characteristics, and generating target recommendation information corresponding to the target service combination.
[0005] In a second aspect, the present disclosure provides a device for generating recommendation information, including: an acquisition module, used to acquire session characteristics of a target object and an initial candidate service information set; a recall module, used to recall a candidate service combination set that meets the session characteristics from the initial candidate service information set using a preset recall rule; an evaluation module, used to evaluate each candidate service information in the candidate service combination set and determine the optimal service parameters of each candidate service information; a recommendation module, used to determine a target service combination from the candidate service combination set according to the optimal service parameters and session characteristics, and generate target recommendation information corresponding to the target service combination.
[0006] In a third aspect, the present disclosure provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for generating recommendation information of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.
[0007] In a fourth aspect, the present disclosure provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for generating recommendation information of the first aspect or any corresponding embodiment thereof.
[0008] In a fifth aspect, the present disclosure provides a computer program product, including computer instructions, where the computer instructions are used to enable a computer to execute the method for generating recommendation information of the first aspect or any corresponding embodiment thereof.
[0009] The method, device, equipment, storage medium and program product for generating recommendation information provided by the present disclosure pre-set recall rules in combination with the mutual exclusion rules between candidate service information, extract session features from the session information when acquiring the session information of the target object, and recall the candidate service combination set that meets the session features from the initial candidate service information set according to the preset recall rules, thereby taking into account the mutual influence between the candidate service information, ensuring that the candidate service combination set can accurately match the session demands of the target object, and improving the accuracy of the candidate service combination set. By evaluating each candidate service information in the candidate service combination set, the optimal service parameters are determined within each candidate service information, and then the target service combination is determined from the candidate service combination set according to the optimal service parameters and session features, thereby combining the internal evaluation of the candidate service information and the combination between different candidate service information to generate the target recommendation information, not only can the optimal service parameters of a single candidate service information be obtained to achieve the best service effect of the single candidate service information, but also the optimal decision of the target service combination can be made based on the optimal service parameters and session features. By generating corresponding target recommendation information according to the service content of the target service combination, it is convenient to flexibly determine the target service combination according to the optimal service parameters and session characteristics, thereby ensuring the flexibility of generating the target recommendation information. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] In order to more clearly illustrate the specific embodiments of the present disclosure or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0011] Figure 1 is a schematic diagram of determining a service information combination according to an embodiment of the present disclosure;
[0012] Figure 2 is a flowchart of a method for generating recommendation information according to an embodiment of the present disclosure;
[0013] Figure 3is a flowchart of another method for generating recommendation information according to an embodiment of the present disclosure;
[0014] Figure 4 is a flowchart of another method for generating recommendation information according to an embodiment of the present disclosure;
[0015] Figure 5 is a reasoning diagram of a compensation service according to an embodiment of the present disclosure;
[0016] Figure 6 is a gain decision schematic diagram according to an embodiment of the present disclosure;
[0017] Figure 7 is a structural block diagram of a device for generating recommendation information according to an embodiment of the present disclosure;
[0018] Figure 8 It is a schematic diagram of the hardware structure of the computer device of the embodiment of the present disclosure. DETAILED DESCRIPTION
[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present disclosure clearer, the technical solution in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments are part of the embodiments of the present disclosure, rather than all the embodiments. Based on the embodiments in the present disclosure, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present disclosure.
[0020] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.
[0021] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can autonomously choose whether to provide personal information to software or hardware such as an electronic device, application, server, or storage medium that performs the operation of the technical solution of the present disclosure according to the prompt message.
[0022] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.
[0023] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation of the present disclosure. Other methods that meet the relevant laws and regulations may also be applied to the implementation of the present disclosure.
[0024] It is understandable that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and relevant provisions.
[0025] At present, the generation of recommendation information is mostly solved on a case-by-case basis for users. Common methods include the following two categories:
[0026] (1) Similar question matching: First, build a "question-solution" knowledge base in a specific field, and then match the user's question Q1 with the question Q2 in the knowledge base. The "knowledge" corresponding to the question Q2 with the highest similarity in the knowledge base is taken as the final recommendation result. However, this method depends on the rationality of the "question-solution" configuration in the knowledge base and whether the granularity can fully cover all scenarios. However, user questions will vary to varying degrees. If there is no completely corresponding "question-solution" configuration in the knowledge base, it is likely that the answer will not be what is asked. In addition, such a knowledge base configuration is pre-set and cannot be well adjusted dynamically and flexibly in real time in combination with other constraints such as changes in the business data of the day in the business scenario.
[0027] (2) Intent recognition - combining intent with other features to recommend information. However, since it introduces other features, it can combine different state features to provide better solutions for the same problem, but it still does not take the dynamic constraints of business data into consideration.
[0028] It can be seen that the above solutions all have a problem, that is, when the business prior knowledge / constraints change, it is necessary to re-accumulate training data, re-train, and deploy, and the optimization cost caused by each change is relatively high. At the same time, since the above two solutions are both solutions that recommend solutions one by one, when users come in and raise multiple questions / demands, the above solutions do not consider the mutual exclusion / superposition enhancement effect between the recommended information, which may lead to the problem of secondary adjustment of the recommended information and cost redundancy.
[0029] Based on this, the disclosed technical solution provides a service information combination to generate recommendation information, thereby providing personalized service information combinations for different user problem demands, thereby bringing users a better demand resolution experience. Figure 1As shown, the system side is deployed with an evaluation model for single service decision and a recommendation model for combined service decision. The system side can receive new messages initiated by the user side, extract the session features corresponding to the new messages, and recall all legal candidate service combinations according to business rule constraints. For each candidate service information in the candidate service combination, the evaluation model is used to make personalized adjustments in each single service information to obtain the single service information decision situation under the optimal gain. Therefore, within the set of candidate service combinations output under the constraints of business rules, based on more sufficient service information and combined with the guidance of business prior knowledge, the mutually exclusive effect or superposition enhancement effect between different service information is considered to decide the best service combination for the user, providing more flexible personalized service combination recommendations.
[0030] According to an embodiment of the present disclosure, an embodiment of a method for generating recommendation information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.
[0031] In this embodiment, a method for generating recommendation information is provided, which can be used in computer devices, such as computers, tablet computers, etc. Figure 2 is a flow chart of a method for generating recommendation information according to an embodiment of the present disclosure. Figure 2 As shown, the process includes the following steps:
[0032] Step S101: Acquire the session characteristics of the target object and an initial candidate service information set.
[0033] Session features are features possessed by session information sent by the target object. Specifically, the session features include static features, historical features, real-time behavior features, and real-time conversation features, among which static features include the object identifier and shopping level of the target object; historical features include order features of the past N days (such as product categories, product prices, etc.), the active dates of the target object, the historical shopping dates of the target object, etc.; real-time behavior features include returns, refunds, and complaints initiated by the target object; real-time conversation features include conversation features between the conversation agent and the target object, conversation features between customer service and the target object, and conversation context, etc.
[0034] Specifically, the conversation information initiated for the target object has a corresponding feature library, which includes all features related to the target object. When the target object initiates new session information, the session information context is analyzed to trigger the acquisition of session features. The computer device can query the session features related to the session information from the feature library.
[0035] The initial candidate service information set is an unordered set of all candidate service information for the target object's conversation behavior, including basic service plans and additional service plans. The basic service plan may include return and refund service, partial refund service, etc., and the additional service plan may include contact method, first contact time limit for work order, etc.
[0036] Specifically, the session message triggered by the target object has a corresponding database, which includes all candidate service information related to the target object. When the target object initiates a new session information, it will trigger the acquisition of the initial candidate service information set. Accordingly, the computer device can access the database to read the initial candidate service information set related to the session message from the database.
[0037] Step S102: recalling a set of candidate service combinations that meet the session characteristics from an initial set of candidate service information using a preset recall rule.
[0038] The preset recall rule is a pre-set rule for recalling candidate services, which is set by using the business characteristics of the business scenario and the mutual exclusion rules between candidate service solutions. The candidate service combination set is a combination set of all legal solutions that meet the session characteristics.
[0039] The state (optional state or prohibited state) of each initial candidate service information in the initial candidate service information set is calculated according to the preset recall rule. If the initial candidate service information is in the prohibited state, it means that it does not meet the session characteristics. The initial candidate service information of all optional states is screened out from the initial candidate service information set according to the preset recall rule. For each initial candidate service information of the optional state, all the candidate service information in the optional state is combined according to the mutually exclusive rule in the preset recall rule to obtain all legal candidate service combination sets.
[0040] Step S103: Evaluate each candidate service information in the candidate service combination set to determine the optimal service parameter of each candidate service information.
[0041] The optimal service parameters are the service parameters when the candidate service information achieves the optimal service effect, such as the optimal first-touch time, the optimal compensation amount, etc. The legal parameter range of the candidate service information is calculated according to the business rules in the candidate service information, and the candidate service information is executed in sequence according to each parameter within the legal parameter range to obtain the service execution results corresponding to each parameter. The service execution results corresponding to each parameter are compared to determine the optimal service execution result, and the service parameters corresponding to the optimal service execution result are determined as the optimal service parameters.
[0042] Step S104: determine a target service combination from the candidate service combination set according to the optimal service parameters and the session characteristics, and generate target recommendation information corresponding to the target service combination.
[0043] The target recommendation information is information generated for the conversation information initiated by the target object at that time, and is used to guide customer service personnel to respond to the conversation information initiated by the target object at that time. Specifically, the question request of the conversation information initiated by the target object at that time can be determined in combination with the conversation characteristics, and the corresponding optimal service parameters are assigned to each candidate service information in each candidate service combination. In this way, according to the principle of "responding to every request" for the target object, the optimal service parameters and the question request corresponding to the conversation characteristics are used to make decisions on each candidate service combination in the candidate service combination set to obtain the optimal target service combination. Then, each service information in the target service combination is parsed, and the target recommendation information is generated according to the service information content.
[0044] The information recommendation method provided in this embodiment pre-sets a recall rule in combination with the mutual exclusion rule between the candidate service information. When the session information of the target object is obtained, the session features are extracted from the session information, and the candidate service combination set that meets the session features is recalled from the initial candidate service information set according to the preset recall rule, thereby taking into account the mutual influence between the candidate service information, ensuring that the candidate service combination set can accurately match the session demands of the target object, and improving the accuracy of the candidate service combination set. By evaluating each candidate service information in the candidate service combination set, the optimal service parameters are determined within each candidate service information, and then the target service combination is determined from the candidate service combination set according to the optimal service parameters and the session features, thereby combining the internal evaluation of the candidate service information and the combination of different candidate service information to generate the target recommendation information, not only can the optimal service parameters of a single candidate service information be obtained to achieve the best service effect of the single candidate service information, but also the optimal decision of the target service combination can be made based on the optimal service parameters and the session features. By generating the corresponding target recommendation information according to the service content of the target service combination, it is convenient to flexibly determine the target service combination according to the optimal service parameters and the session features, and the flexibility of generating the target recommendation information is guaranteed.
[0045] In this embodiment, a method for generating recommendation information is provided, which can be used in computer devices, such as computers, tablet computers, etc. Figure 3 is a flow chart of a method for generating recommendation information according to an embodiment of the present disclosure. Figure 3 As shown, the process includes the following steps:
[0046] Step S201, obtaining the session characteristics of the target object and an initial candidate service information set. For details, please refer to the description of the corresponding steps in the above embodiment, which will not be repeated here.
[0047] Step S202: recall a set of candidate service combinations that meet the session characteristics from an initial set of candidate service information using a preset recall rule.
[0048] Specifically, the above step S202 includes:
[0049] Step S2021, determining the state of each initial candidate service information in the initial candidate service information set according to the session characteristics, the initial candidate service information has a prohibited state and an optional state.
[0050] The prohibited state indicates that the initial candidate service information does not match the session characteristics; the optional state indicates that the initial candidate service information matches the session characteristics. The session characteristics are analyzed to determine the target object's intention to initiate the current session information, and the candidate service information related to the current intention is determined. Then, each initial candidate service information in the initial candidate service information set is evaluated in combination with the current intention and the session characteristics, and a prohibited state or an optional state is generated for each initial candidate service information.
[0051] Step S2022: Filter the first initial candidate service information in the prohibited state in the initial candidate service information set to obtain a plurality of second initial candidate service information in the selectable state.
[0052] Since the prohibited state indicates that the candidate service information does not match the session characteristics of the current session information, that is, the candidate service information cannot be used as a candidate service solution for the session characteristics. At this time, the state label of each initial candidate service information in the initial candidate service information set is detected, and the first initial candidate service information with the state label of the prohibited state in the initial candidate service information set is filtered out to obtain the second initial candidate service information in the optional state.
[0053] Step S2023: combine the plurality of second initial candidate service information using a preset recall rule to obtain a candidate service combination set.
[0054] As described above, the preset recall rules are used to recall candidate service combinations. Due to the mutually exclusive rules of service combinations, different second initial candidate service information may not be combined, such as the return and refund service cannot be combined with the partial refund service. Therefore, when multiple second initial candidate service information in an optional state is obtained, the mutually exclusive rules in the preset recall rules can be used to combine each second initial candidate service information to obtain multiple legal candidate service combinations that meet the business scenarios, and the multiple candidate service combinations constitute a candidate service combination set.
[0055] Step S203: Evaluate each candidate service information in the candidate service combination set to determine the optimal service parameter of each candidate service information. Please refer to the description of the corresponding steps in the above embodiment for details, which will not be repeated here.
[0056] Step S204: determine the target service combination from the candidate service combination set according to the optimal service parameters and session characteristics, and generate target recommendation information corresponding to the target service combination. For details, please refer to the relevant description of the corresponding steps in the above embodiment, which will not be repeated here.
[0057] The information recommendation method provided in this embodiment uses the conversation features as a reference to determine the first initial candidate service information in a prohibited state and the second initial candidate service information in an optional state in the initial candidate service information set, and then combines the second initial candidate service information according to the preset recall rule to obtain a candidate service combination set, so as to determine an accurate candidate service combination set in combination with the conversation features of the target object, and ensure that the candidate service combination set obtained under the constraints of the preset recall rules can cover the conversation features of the target object, so that the candidate service combination set can have more sufficient information, which is convenient for subsequent decision-making on the target service combination.
[0058] In this embodiment, a method for generating recommendation information is provided, which can be used in computer devices, such as computers, tablet computers, etc. Figure 4 is a flow chart of a method for generating recommendation information according to an embodiment of the present disclosure. Figure 4 As shown, the process includes the following steps:
[0059] Step S301: Acquire the session characteristics of the target object and the initial candidate service information set. Please refer to the description of the corresponding steps in the above embodiment for details, which will not be repeated here.
[0060] Step S302: recall a set of candidate service combinations that meet the session characteristics from the initial candidate service information set using a preset recall rule. For details, please refer to the description of the corresponding steps in the above embodiment, which will not be repeated here.
[0061] Step S303: Evaluate each candidate service information in the candidate service combination set to determine the optimal service parameter of each candidate service information.
[0062] Specifically, the above step S303 includes:
[0063] Step S3031: Obtain constraint rules for candidate service information.
[0064] Constraint rules are used to constrain each candidate service information that is in an optional state. Specifically, different candidate service information corresponds to different constraint rules, and different business scenarios correspond to different constraint rules. When the candidate service combination set is recalled from the initial candidate service information set, an evaluation decision for each candidate service information is triggered. At this time, the constraint factors affecting the candidate service information can be determined according to the business scenario (such as the specific needs of the target object, the business rules corresponding to the business scenario, the budget limit of the candidate service, etc.), and the corresponding constraint rules are determined according to the constraint factors of the candidate service information. In a specific example, if the session information initiated by the target object includes the requirement of "quick response", the constraint rule can be that the candidate service must be able to respond within X time.
[0065] Step S3032: predicting the service parameters of each candidate service information according to the constraint rules, and generating a parameter prediction range corresponding to each candidate service information.
[0066] The parameter prediction range is the legal parameter range of the candidate service information. Specifically, the pre-trained decision model is used to simulate the process of executing each candidate service information according to the service parameters, and the parameter prediction range of the candidate service information is determined according to the simulation results of the candidate service information. For example, if the candidate service information is a partial refund service, the constraint rules corresponding to the candidate service information are input into the decision model, so that the decision model predicts the refund parameters of the partial refund service according to the constraint rules, and outputs the corresponding refund parameter prediction range [Q1, Q2].
[0067] Among them, the decision model is trained based on the model architecture of the causal forest model. The decision model simulates the potential execution results of the candidate service information by establishing a large number of decision trees, and performs causal inference on the service parameters of the candidate service information by comparing the average prediction results of the treatment group and the control group, and obtains the parameter prediction range corresponding to the candidate service information.
[0068] Step S3033: Evaluate the candidate service information based on the parameter prediction range, and determine the optimal service parameters from the parameter prediction range according to the evaluation result.
[0069] According to the parameter prediction range, the candidate service information is simulated and executed in order of service parameters from small to large, and the simulation execution results corresponding to each service parameter are obtained. According to each simulation execution result, the execution effect of the candidate service information is evaluated, the optimal execution result is determined, and the service parameter corresponding to the optimal execution result is determined as the optimal service parameter.
[0070] In some optional implementations, the above step S3033 includes:
[0071] Step a1: for any current parameter in the parameter prediction range, determine the first service data generated by the candidate service information under the current parameter.
[0072] Step a2: If the first service data meets the constraint threshold, the current parameter is expanded within the parameter prediction range to determine the second service data generated after the current parameter is expanded.
[0073] Step a3: if the second service data meets the constraint threshold, the first service data and the second service data are compared, candidate service information is evaluated based on the service data comparison result, and the optimal service parameters are determined from the parameter prediction range.
[0074] Step a4: if the first service data meets the constraint threshold and the second service data does not meet the constraint threshold, the service parameter corresponding to the first service data is determined as the optimal service parameter.
[0075] The first service data is the simulated data generated by executing the candidate service according to the current parameters, and the effect of executing the candidate service according to the current parameters is characterized by the first service data. The second service data is the simulated data generated by executing the candidate service after expanding the current parameters, and is used to characterize the effect of executing the candidate service after expanding the current parameters. The constraint threshold is a pre-set parameter indicator value for whether the candidate service information meets the business rules.
[0076] The execution of the candidate service is simulated according to the current parameters through the decision model to generate the corresponding first service data. If the first service data meets the constraint threshold, the current parameters are further expanded within the parameter prediction range, and the execution of the candidate service is simulated according to the expanded parameters through the decision model to generate the corresponding second service data. If the second service data meets the constraint threshold, the first service data and the second service data are compared to determine the comparison result between the first service data and the second service data. If it is determined based on the service data comparison result that the second service data is better than the first service data, the service parameters corresponding to the second service data can be temporarily set as the optimal service parameters, and then the parameters are further expanded within the parameter prediction range to obtain the corresponding service data, and the service data is compared with the second service data. If it is determined that the service data is better than the second service data, the optimal service parameters are updated with the service parameters that are expanded again. By analogy, the candidate service information is evaluated according to the comparison results between the service data, and the optimal service parameters of the candidate service information can be determined from the parameter prediction range.
[0077] In the above implementation, when it is determined that the service data generated by executing the candidate service information according to the current parameters meets the constraint threshold, the current parameters are further expanded according to the parameter prediction range, and it is determined whether the service data generated by executing the candidate service information according to the expanded current parameters meets the corresponding constraint threshold. If the constraint threshold is met, the optimal service parameters are determined by comparing the service data corresponding to different service parameters; if the constraint threshold is not met, all service data that meet the constraint threshold can be compared to determine the optimal service parameters based on the comparison results. Thus, personalized adjustment of service parameters is performed in each candidate service information according to the parameter prediction range, so as to make personalized adjustment decisions in combination with the conversation characteristics of the target object, thereby achieving the best service effect of a single candidate service information.
[0078] In some optional implementations, the above method further includes:
[0079] Step b1, obtaining the remaining parameter range except the current parameter.
[0080] Step b2: determining the constraint threshold corresponding to each service parameter within the remaining parameter range according to the mapping relationship between the service parameter and the constraint threshold.
[0081] The remaining parameter range is the parameter interval excluding the current parameter within the parameter prediction range. The mapping relationship can be represented by a mapping table between the remaining parameters and the constraint thresholds. Whenever the service parameters are updated, the constraint thresholds corresponding to the service parameters are updated according to the mapping relationship. Among them, the constraint thresholds corresponding to each service parameter are calculated using the service data generated by executing the candidate service information. The specific calculation method can be determined according to the actual business scenario and is not limited here.
[0082] In the above implementation, the constraint thresholds corresponding to each service parameter are dynamically adjusted in combination with the mapping relationship between the service parameters and the constraint thresholds, thereby achieving the best decision on the optimal service parameters for a single candidate service information, ensuring the overall consistency of the service data, and improving service efficiency.
[0083] In some optional implementations, the above method further includes:
[0084] Step c1, obtaining the fuse rule for the candidate service information.
[0085] Step c2: If the execution result of the candidate service information triggers the fuse rule, the optimal service parameters are determined according to the fuse rule.
[0086] The fuse rule is a fallback rule for the candidate service information to prevent the service data generated by executing the candidate service information from exceeding the limit. When the candidate service information is executed according to the service parameters in the parameter prediction range, the service data generated by executing the candidate service information is compared with the fuse rule to determine whether the service data generated by executing the candidate service information triggers the fuse rule.
[0087] If the service data generated by executing the candidate service information triggers the fuse rule, it means that the service data generated for the candidate service information has exceeded the limit. At this time, the optimal service parameter of the candidate service information is set to the minimum value according to the fuse rule.
[0088] In the above implementation, by setting the fuse rule, when the execution result of the candidate service information triggers the fuse rule, the optimal service parameters are determined according to the fuse rule, thereby avoiding the service data of the candidate service information from exceeding the limit and ensuring the implementation effect of the candidate service information.
[0089] In a specific example, taking the candidate service information as compensation service and the customer service determining the compensation amount as an example, at this time, we can consider deciding the final output result based on the gain decision model (Uplift Model) by predicting the gain of different plans on business indicators (such as FCR), so that the gain decision model has a certain result interpretation ability, such as compensation of yy yuan, or 0.5 hours / 2 hours of first contact with the target object.
[0090] Specifically, the decision-making framework for claims services is composed of "business rules + gain decision model", among which the business rules are responsible for bottom-line risk constraints to ensure that the subsequent gain decision model will not "overstep the boundary"; the gain decision model is within the constraints of business rules, using richer features and more powerful model decision-making capabilities to maximize service effects.
[0091] When the customer service triggers the compensation tool for the target user and his / her order that sent the session information, the intelligent compensation link process is as follows: Figure 5 As shown in the figure, when the customer service agent triggers the compensation tool, the conversation features of the target object are queried from the feature library. The customer service agent can select the compensation scenario and compensation reason, and impose risk constraints on the current compensation according to the business rules, determine the constraint rules of the compensation service (i.e., whether compensation is available) and the parameter prediction range corresponding to the constraint rules (i.e., the upper and lower limits of compensation). The execution results of the compensation service are evaluated according to the constraint rules and parameter prediction range through the gain decision model. The business indicator gains (such as money-effectiveness value) of different compensation plans are different. The compensation plan and business indicator gains are predicted according to the gain maximization, and the prediction results are quantified to produce compensation-sensitive scoring labels for the target object.
[0092] Then, the cost constraint rule is used to balance the cost and gain to obtain the recommended compensation amount. In the process of determining the compensation amount, the recommendation link of the compensation amount is optimized by increasing the cache and feature pre-fetching, reducing the prediction time. Among them, the gain decision model uses the causal forest model as the basic model selection, uses the characteristics of random forests, simulates potential results by establishing a large number of decision trees, and estimates the causal effect by comparing the average prediction results of the treatment group and the control group.
[0093] Since the online decision-making of intelligent compensation needs to consider preference indicators such as FCR and satisfaction, it is also necessary to consider the limitations of funds, costs, traffic, etc., so as to maximize the overall ROI under the premise of limited resources. That is, after the gain decision model calculates the intervention gain, the compensation strategy under the cost constraint rule mainly considers two aspects: controlling the compensation cost not to exceed the budget and controlling the compensation budget to be fully used. At this time, the greedy compensation strategy and dynamic money-effectiveness threshold can be used to calculate the current optimal compensation plan, such as Figure 6 The specific calculation method is as follows:
[0094] Index gain FCR = f(x, treatment_i) - f(x, treatment_none);
[0095] Optimal money efficiency = index gain FCR / compensation amount treatment_i.
[0096] Among them, f(x,treatment_i) represents the historical payment amount at the xth hour, and f(x,treatment_none) means that no payment will be made for the target object x.
[0097] Specifically, the money-effectiveness threshold T must ensure that all claims are paid within the budget and that the final total compensation amount does not exceed the total budget setting excessively. Calculate the optimal money-effectiveness threshold T for the next hour at the hourly level, and then the compensation amount budget B for the next hour is i The calculation is as follows:
[0098] B i = (the amount of compensation paid in the xth hour in history / the amount of compensation paid in the xth to 24th hour in history) * (the total cost of compensation paid on the day - the total amount of compensation paid)
[0099] The money efficiency threshold T for the next hour i The calculation is as follows:
[0100] T i = The historical cost limit for the xth hour is B i The corresponding T i '
[0101] At the same time, the circuit breaker rules are set. When the total amount of successful compensation on the day exceeds the circuit breaker upper limit, the compensation process will be circuit breaker and the compensation amount corresponding to the compensation service will be set to 0, that is, compensation is not recommended.
[0102] It should be noted that the historical average hourly compensation amount and the total amount of compensation paid on the day are counted and obtained online at regular intervals. i With T i ′ to avoid time-consuming real-time recalculation during online use.
[0103] Step S304: determine a target service combination from the candidate service combination set according to the optimal service parameters and the session characteristics, and generate target recommendation information corresponding to the target service combination.
[0104] Specifically, the above step S304 includes:
[0105] Step S3041: extract prompt structure information from the optimal service parameters and session features, and generate combined prompt description information according to the prompt structure information.
[0106] Hint structure information refers to a series of information or instructions provided to the recommendation model, which helps the recommendation model understand the requirements of the recommendation task, the expected output format, and the processing of input data. The recommendation model can be trained based on the model architecture of a large language model or the model architecture of a machine learning model. There is no specific limitation here, as long as it can achieve the recommendation task.
[0107] Specifically, the prompt structure information may include task background, current status, and task description. The task background describes the combined information and reasoning process logic from an overall perspective; the current status describes all the status information and conversation information of the current object, order, and merchant; the task description is used to disassemble the combined information recommendation task, which includes: extracting and identifying "questions in conversation information" and "demands of the target object", recalling a reasonable set of recommended candidate service combinations based on the current status and the above recognition results, and recommending candidate service combinations based on the current status and the recall results.
[0108] The prompt structure information such as task background, current status and task description is extracted from the optimal service parameters and session features respectively, and the prompt structure information is described in natural language to form combined prompt description information so as to clearly describe the chain recommendation task.
[0109] Step S3042: Determine a target service combination from the candidate service combination set using the combination prompt description information.
[0110] According to the combination prompt description information, the recommendation model is guided to gradually think about the knowledge correlation between the session characteristics and each candidate service, and the optimal candidate service combination that meets the business needs is inferred based on the knowledge correlation, and the optimal candidate service combination is determined as the target service combination.
[0111] Step S3043: Generate target recommendation information corresponding to the target service combination based on the combination information of the target service combination.
[0112] The combination information is used to represent the integration or combination of different services. The multiple services included in the target service combination are determined according to the combination information. Different services have different service contents. By combining the service contents of each service in the target service combination, the target recommendation information composed of different service contents can be obtained, and the recommendation logic explanation can be generated in text form.
[0113] In some optional implementations, the above method further includes:
[0114] Step d1, simplifying the combined prompt description information based on the conversation characteristics to obtain simplified combined prompt description information.
[0115] Step d2: determining a target service combination from a set of candidate service combinations according to the simplified combination prompt description information.
[0116] According to the current conversation characteristics and the specific recalled candidate service information, the core elements of the combined prompt description information are determined, and redundant information is removed to simplify the combined prompt description information, so that the simplified combined prompt description information is more direct and specific, ensuring that the combined prompt description information is highly matched with the demands of the target object.
[0117] According to the simplified combination prompt description information, the recommendation model is guided to think about the knowledge correlation between the session features and each candidate service, and the knowledge correlation is used to infer the optimal candidate service combination, and the optimal candidate service combination obtained by reasoning is determined as the target service combination. In this way, the reasoning effect of the recommendation model can be better optimized, which is convenient for providing the target object with recommendation information that is closer to the needs, improves the efficiency of generating recommendation information, and thus improves service efficiency.
[0118] The information recommendation method provided in this embodiment combines the prediction ability of the decision model for the optimal service parameters and the logical reasoning ability of the recommendation model, so that after obtaining relatively accurate gain information, the relevant knowledge of the business scenario is added to the decision process of the target service combination through the prompt engineering of the recommendation model, thereby improving the decision accuracy of the target service combination. By predicting each candidate service information according to the constraint rules, the service parameters of the candidate service information are dynamically adjusted within the parameter prediction range to achieve the optimal control effect for the candidate service information. At the same time, the relevant knowledge of the business scenario is injected into the prompt structure information in the form of natural language. When the relevant knowledge of the business scenario changes, the recommendation model can respond more flexibly to determine the best target service combination, and realize the flexible adjustment of the target service combination. Therefore, the personalized combination of service information can be performed according to the conversation characteristics of the target object, ensuring that the target service combination can match the conversation characteristics of the target object, greatly improving the accuracy of the target service combination, thereby improving the generation accuracy of the target recommendation information, avoiding the query of service information one by one, saving the generation time of the recommendation information, and improving the service efficiency for the conversation demands of the target object.
[0119] In this embodiment, a device for generating recommendation information is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.
[0120] This embodiment provides a device for generating recommendation information, such as Figure 7 As shown, including:
[0121] The acquisition module 401 is used to acquire the session characteristics of the target object and an initial candidate service information set.
[0122] The recall module 402 is used to recall a set of candidate service combinations that meet the session characteristics from the initial candidate service information set using a preset recall rule.
[0123] The evaluation module 403 is used to evaluate each candidate service information in the candidate service combination set and determine the optimal service parameter of each candidate service information.
[0124] The recommendation module 404 is used to determine a target service combination from the candidate service combination set according to the optimal service parameters and session characteristics, and generate target recommendation information corresponding to the target service combination.
[0125] In some optional implementations, the recall module 402 includes:
[0126] The state determination unit is used to determine the state of each initial candidate service information in the initial candidate service information set according to the session characteristics, and the initial candidate service information has a prohibited state and an optional state.
[0127] The filtering unit is used to filter the first initial candidate service information in a prohibited state in the initial candidate service information set to obtain a plurality of second initial candidate service information in an optional state.
[0128] The combining unit is used to combine a plurality of second initial candidate service information by using a preset recall rule to obtain a candidate service combination set.
[0129] In some optional implementations, the evaluation module 403 includes:
[0130] The rule acquisition unit is used to acquire constraint rules for candidate service information.
[0131] The parameter prediction unit is used to predict the service parameters of each candidate service information according to the constraint rules, and generate the parameter prediction range corresponding to each candidate service information.
[0132] The evaluation unit is used to evaluate the candidate service information based on the parameter prediction range, and determine the optimal service parameters from the parameter prediction range according to the evaluation result.
[0133] In some optional implementations, the evaluation unit includes:
[0134] The first data generating subunit is used to determine, for any current parameter in the parameter prediction range, first service data generated by the candidate service information under the current parameter.
[0135] The second data generating subunit is used to expand the current parameters within the parameter prediction range if the first service data meets the constraint threshold, and determine the second service data generated after the current parameters are expanded.
[0136] The first service parameter determination unit is used to compare the first service data with the second service data if the second service data meets the constraint threshold, evaluate the candidate service information based on the service data comparison result, and determine the optimal service parameter from the parameter prediction range.
[0137] In some optional implementations, the evaluation unit further includes:
[0138] The second service parameter determining unit is configured to determine the service parameter corresponding to the first service data as the optimal service parameter if the first service data meets the constraint threshold and the second service data does not meet the constraint threshold.
[0139] In some optional implementations, the evaluation unit further includes:
[0140] The parameter range acquisition subunit is used to acquire the remaining parameter range except the current parameter.
[0141] The constraint threshold determination subunit is used to determine the constraint threshold corresponding to each service parameter within the remaining parameter range according to the mapping relationship between the service parameter and the constraint threshold.
[0142] In some optional implementations, the evaluation unit further includes:
[0143] The circuit breaker rule acquisition subunit is used to obtain the circuit breaker rules for the candidate service information.
[0144] The third service parameter determination unit is used to determine the optimal service parameter according to the fuse rule if the execution result of the candidate service information triggers the fuse rule.
[0145] In some optional implementations, the recommendation module 404 includes:
[0146] The prompt generating unit is used to extract the prompt structure information from the optimal service parameter and the session feature, and generate the combined prompt description information according to the prompt structure information.
[0147] The service combination determining unit is used to determine a target service combination from a set of candidate service combinations by using the combination prompt description information.
[0148] The first recommendation information generating unit is configured to generate target recommendation information corresponding to the target service combination based on the combination information of the target service combination.
[0149] In some optional implementations, the recommendation module 404 further includes:
[0150] The information simplification unit is used to simplify the combined prompt description information based on the conversation feature to obtain simplified combined prompt description information.
[0151] The second recommendation information generating unit is used to determine the target service combination from the candidate service combination set according to the simplified combination prompt description information.
[0152] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.
[0153] The recommendation information generating device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.
[0154] The device for generating recommendation information provided in this embodiment pre-sets a recall rule in combination with the mutual exclusion rule between candidate service information. When the session information of the target object is obtained, the session features are extracted from the session information, and the candidate service combination set that meets the session features is recalled from the initial candidate service information set according to the preset recall rule, thereby taking into account the mutual influence between the candidate service information, ensuring that the candidate service combination set can accurately match the session demands of the target object, and improving the accuracy of the candidate service combination set. By evaluating each candidate service information in the candidate service combination set, the optimal service parameters are determined within each candidate service information, and then the target service combination is determined from the candidate service combination set according to the optimal service parameters and the session features, thereby combining the internal evaluation of the candidate service information and the combination between different candidate service information to generate the target recommendation information, not only can the optimal service parameters of a single candidate service information be obtained to achieve the best service effect of the single candidate service information, but also the optimal decision of the target service combination can be made based on the optimal service parameters and the session features. By generating the corresponding target recommendation information according to the service content of the target service combination, it is convenient to flexibly determine the target service combination according to the optimal service parameters and the session features, and the flexibility of generating the target recommendation information is guaranteed.
[0155] The present disclosure also provides a computer device having the above Figure 7 The device for generating the recommendation information shown.
[0156] See also Figure 8 , Figure 8 is a schematic diagram of a computer device provided by an optional embodiment of the present disclosure, such as Figure 8 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 8 A processor 10 is taken as an example.
[0157] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.
[0158] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.
[0159] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store data created according to the use of the computer device, and the like.
[0160] In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.
[0161] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.
[0162] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 8 The example of connecting through bus is taken in the following.
[0163] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.
[0164] The computer device also includes a communication interface, which is used for the computer device to communicate with other devices or a communication network.
[0165] The embodiments of the present disclosure also provide a computer-readable storage medium. The above-mentioned method according to the embodiments of the present disclosure can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium and downloaded through a network, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.
[0166] A part of the present disclosure may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present disclosure through the operation of the computer. Those skilled in the art should understand that the existence of computer program instructions in computer-readable media includes, but is not limited to, source files, executable files, installation package files, etc., and accordingly, the way in which computer program instructions are executed by a computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to the computer.
[0167] Although the embodiments of the present disclosure have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present disclosure, and such modifications and variations are all within the scope defined by the appended claims.
Claims
1. A method for generating recommendation information, characterized in that: The method comprises: Obtaining the session characteristics of the target object and an initial candidate service information set; Recalling a set of candidate service combinations that meet the session characteristics from the initial candidate service information set using a preset recall rule; Evaluate each piece of candidate service information in the candidate service combination set to determine the optimal service parameter of each piece of candidate service information; According to the optimal service parameters and the session characteristics, a target service combination is determined from the candidate service combination set, and target recommendation information corresponding to the target service combination is generated.
2. The method according to claim 1, characterized in that The step of using a preset recall rule to recall a candidate service combination set that meets the session characteristics from the initial candidate service information set includes: Determine the state of each piece of initial candidate service information in the initial candidate service information set according to the session feature, wherein the initial candidate service information has a prohibited state and an optional state; filtering the first initial candidate service information in the prohibited state in the initial candidate service information set to obtain a plurality of second initial candidate service information in an optional state; The plurality of the second initial candidate service information are combined using a preset recall rule to obtain the candidate service combination set.
3. The method according to claim 1, characterized in that The step of evaluating each candidate service information in the candidate service combination set to determine the optimal service parameter of each candidate service information includes: Obtaining constraint rules for the candidate service information; Predicting the service parameters of each of the candidate service information according to the constraint rules, and generating parameter prediction ranges corresponding to each of the candidate service information; The candidate service information is evaluated based on the parameter prediction range, and the optimal service parameter is determined from the parameter prediction range according to the evaluation result.
4. The method according to claim 3, characterized in that The candidate service information is evaluated based on the parameter prediction range, and the optimal service parameter is determined from the parameter prediction range according to the evaluation result, including: For any current parameter in the parameter prediction range, determining first service data generated by the candidate service information under the current parameter; If the first service data satisfies the constraint threshold, expanding the current parameter within the parameter prediction range, and determining the second service data generated after the current parameter is expanded; If the second service data meets the constraint threshold, the first service data and the second service data are compared, the candidate service information is evaluated based on the service data comparison result, and the optimal service parameter is determined from the parameter prediction range.
5. The method according to claim 4, characterized in that Also includes: If the first service data satisfies the constraint threshold and the second service data does not satisfy the constraint threshold, the service parameter corresponding to the first service data is determined as the optimal service parameter.
6. The method according to claim 4, characterized in that Also includes: Obtain the remaining parameter ranges except the current parameter; According to the mapping relationship between the service parameters and the constraint thresholds, the constraint thresholds corresponding to the respective service parameters within the remaining parameter range are determined.
7. The method according to claim 3, characterized in that Also includes: Obtaining a fuse rule for the candidate service information; If the execution result of the candidate service information triggers the fuse rule, the optimal service parameter is determined according to the fuse rule.
8. The method according to claim 1, characterized in that The step of determining a target service combination from the candidate service combination set according to the optimal service parameters and the session characteristics, and generating target recommendation information corresponding to the target service combination includes: Extracting prompt structure information from the optimal service parameters and the session characteristics, and generating combined prompt description information according to the prompt structure information; Determine a target service combination from the candidate service combination set using the combination prompt description information; Based on the combination information of the target service combination, target recommendation information corresponding to the target service combination is generated.
9. The method according to claim 8, characterized in that Also includes: Simplifying the combined prompt description information based on the session feature to obtain simplified combined prompt description information; According to the simplified combination prompt description information, a target service combination is determined from the candidate service combination set.
10. A device for generating recommendation information, characterized in that: The device comprises: An acquisition module, used to acquire the session characteristics of the target object and an initial candidate service information set; A recall module, configured to recall a candidate service combination set satisfying the session characteristics from the initial candidate service information set using a preset recall rule; An evaluation module, used to evaluate each candidate service information in the candidate service combination set, and determine the optimal service parameter of each candidate service information; The recommendation module is used to determine a target service combination from the candidate service combination set according to the optimal service parameters and the session characteristics, and generate target recommendation information corresponding to the target service combination.
11. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the method for generating recommendation information according to any one of claims 1 to 9 by executing the computer instructions.
12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the method for generating recommendation information according to any one of claims 1 to 9.
13. A computer program product, characterized in that The invention comprises computer instructions, wherein the computer instructions are used to cause a computer to execute the method for generating recommendation information according to any one of claims 1 to 9.