Event influence degree evaluation method, device and equipment

Abstract

The embodiment of the invention discloses an event influence degree evaluation method, device and equipment. According to the scheme, the method comprises the steps of generating a target cue word by combining event information of a target event related to a target service and target instruction information used for indicating the influence degree of an evaluation event on the target service, the target instruction information carries target instance information reflecting the historical influence degree of the specified event on the target business, so that the large language model can generate an evaluation result reflecting the influence degree of the target event on the target business according to the target cue word.

Description

Technical Field

[0001] The present application relates to the field of large model technology, and in particular to an event impact assessment method, device and equipment. Background Art

[0002] With the popularization of the global Internet and the development of information technology, a huge amount of information can be quickly disseminated, and the process of people obtaining information has become very convenient, making information an important resource involved in the business operation process. However, since it often requires very professional interpretation ability to understand the connection between various events and various businesses provided by the enterprise, ordinary users usually do not have strong information analysis capabilities, and thus cannot accurately interpret and quantitatively evaluate the impact of current events on subsequent business operations, resulting in their inability to take corresponding business processing actions in a timely manner.

[0003] Based on this, how to accurately and conveniently evaluate the impact of various events on business operations has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] The embodiments of this specification provide an event impact assessment method, device, and equipment that can accurately and conveniently assess the impact of various events on the business operation of a target business.

[0005] To solve the above technical problems, the embodiments of this specification are implemented as follows:

[0006] An event impact assessment method provided in an embodiment of the present specification includes:

[0007] Obtaining event information of a target event related to a target business;

[0008] Acquire target instruction information for indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business;

[0009] generating a target prompt word according to the target instruction information and the event information of the target event;

[0010] By using a large language model, an evaluation result reflecting the impact of the target event on the target business is generated according to the target prompt word.

[0011] An event impact assessment device provided in an embodiment of the present specification includes:

[0012] A first acquisition module, used to acquire event information of a target event related to a target business;

[0013] A second acquisition module is used to acquire target instruction information indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business;

[0014] A first generating module, configured to generate a target prompt word according to the target instruction information and the event information of the target event;

[0015] The second generating module is used to generate an evaluation result reflecting the impact of the target event on the target business according to the target prompt word by using a large language model.

[0016] An event impact assessment device provided in an embodiment of the present specification includes:

[0017] at least one processor; and,

[0018] a memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0020] Obtaining event information of a target event related to a target business;

[0021] Acquire target instruction information for indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business;

[0022] generating a target prompt word according to the target instruction information and the event information of the target event;

[0023] By using a large language model, an evaluation result reflecting the impact of the target event on the target business is generated according to the target prompt word.

[0024] At least one embodiment provided in this specification can achieve the following beneficial effects:

[0025] By combining the event information of the target event related to the target business and the target instruction information used to indicate the impact of the evaluation event on the target business, a target prompt word is generated, and the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business. Based on the super natural language understanding and natural language generation capabilities of the large language model, the large language model can better understand and infer the meaning of the input target prompt word, so as to conveniently generate an evaluation result that accurately reflects the impact of the target event on the target business, and realize an accurate quantitative evaluation of the relationship and impact between the target event and the operation of the target business. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0027] Figure 1 A schematic diagram of an application scenario of an event impact assessment method provided in an embodiment of this specification;

[0028] Figure 2 A flowchart of an event impact assessment method provided in an embodiment of this specification;

[0029] Figure 3 The embodiments of this specification provide corresponding to Figure 2 A swimlane flow diagram of the event impact assessment method in ;

[0030] Figure 4 The embodiments of this specification provide corresponding to Figure 2 A schematic diagram of the structure of an event impact assessment device;

[0031] Figure 5 The embodiments of this specification provide corresponding to Figure 2 A structural schematic diagram of an event impact assessment device. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of one or more embodiments of this specification clearer, the technical solutions of one or more embodiments of this specification will be clearly and completely described below in combination with the specific embodiments of this specification and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of one or more embodiments of this specification.

[0033] The technical solutions provided by the embodiments of this specification are described in detail below in conjunction with the accompanying drawings.

[0034] In the existing technology, the international economic environment, technological breakthroughs, public opinion events, etc. usually affect users' business processing needs. However, since most users have weak ability to interpret these events and lack the ability to quantitatively evaluate the relationship between various events and business operations, users are unable to take corresponding business processing actions in a timely manner, which in turn affects their benefits and rights.

[0035] At present, service providers usually have their staff judge the impact of various events that may occur on the target business they provide based on their personal experience. This method relies heavily on manual experience and has poor accuracy and convenience. In addition, some service providers have begun to use labeled data (positive or negative impact) to train a classification model, so as to use the classification model to simply predict whether a certain event will be beneficial to the business operation. However, this method usually cannot sort and grade the impact of different events on the business operation, and thus cannot meet the current demand for accurate and convenient evaluation of the impact of various events on the business operation.

[0036] In order to solve the defects in the prior art, this solution provides the following embodiments:

[0037] Figure 1 A schematic diagram of an application scenario of an event impact assessment method provided in an embodiment of this specification.

[0038] like Figure 1As shown, in the process of providing target services to users, service providers often use device 101 to collect event information of target events related to the target services. If it is necessary to evaluate the relationship between the target event and the operation of the target service, the device 101 can obtain target instruction information indicating the impact of the evaluation event on the target service from the preset instruction information set in the database 102, and the target instruction information usually carries target instance information reflecting the historical impact of the specified event on the target service. By generating a target prompt word based on the target instruction information and the event information of the target event; by using a large language model, an evaluation result reflecting the impact of the target event on the target service is generated based on the target prompt word, thereby achieving an accurate quantitative evaluation of the relationship and impact between the target event and the operation of the target service.

[0039] In addition, the device 101 can also perform usability verification on the evaluation result. If the usability verification result indicates that the event information of the target event can be recommended to the user group as user recommendation content related to the target business, the target events can be sorted according to the influence in the evaluation result, so as to determine the priority of using the event information of each target event as user recommendation content related to the target business according to the sorting of each target event in the target event sequence. Then, according to the priority corresponding to each target event, the user recommendation content generated based on at least part of the event information of the target event is sent to the user device 103 to assist the user in processing the target business.

[0040] Subsequently, device 101 may also determine a validity verification result for the evaluation result based on a preset feedback index value of the user group for the user recommended content, thereby combining the validity verification result for the evaluation result with the above-mentioned usability verification result to determine the benefit value of the target instruction information used in the impact evaluation process.

[0041] Based on the revenue value of the target instruction information, the target instruction information and the event information of the target event, a reinforcement learning training sample is generated to expand the training sample set. Also, using the reinforcement learning algorithm and the training sample set, a target strategy network model that can be used to generate optimized instruction information is trained, thereby optimizing the preset instruction information set using the target strategy network model. Subsequently, the optimized preset instruction information set can also be used to evaluate the impact of various events on the target business, so as to improve the accuracy and effectiveness of the event impact evaluation scheme.

[0042] The event impact assessment scheme provided in the embodiments of this specification can, on the one hand, use a large language model to interpret target events related to the target business that occur every day according to preset instruction information that carries instance information reflecting the historical impact of the specified event on the target business, so as to accurately quantify and assess the impact of each target event on the target business. On the other hand, by sorting and grading each target event according to its impact on the target business, and giving priority to sending event information of target events with higher impact as user recommendation content to the user group, the preset instruction information required for event impact assessment is optimized in combination with user feedback and reinforcement learning ability, so as to continuously improve the accuracy and effectiveness of the event impact assessment scheme, which is practical.

[0043] Next, an event impact assessment method provided in an embodiment of the specification will be specifically described with reference to the accompanying drawings:

[0044] Figure 2 The following is a flow chart of an event impact assessment method provided in an embodiment of the present specification. From a program perspective, the execution subject of the process can be a device for performing event impact assessment, or an application for performing event impact assessment installed on the device. Figure 2 As shown, the process may include the following steps:

[0045] Step 202: Acquire event information of a target event related to a target business.

[0046] In the embodiments of this specification, the target business may refer to services, products, etc. provided by a service provider to a user. In actual applications, there may be many types of target businesses. For example, the target business may be an insurance business, a resource savings business, a transaction business for target goods / services, a public welfare project, etc., and no specific limitation is made to this.

[0047] The target event may refer to an event that will affect the operation of the target business. Usually, the target event may be an event that occurred in the recent period of time, or may be an event that has occurred or has not occurred and is manually set. In practical applications, there may be many types of target events, such as changes in domestic and foreign economic policies, technological breakthroughs, public opinion events, etc., and there is no specific limitation on this.

[0048] In practical applications, Figure 2There are many ways for the execution subject of the method to obtain the event information of the target event. For example, news information crawled from the Internet can be used as the event information of the target event, or change information of the service provider's operation and control rules for the target business can be used as the event information of the target event, or business data of a specified type purchased from a data provider can be used as the event information of the target event. There is no specific limitation on this.

[0049] Step 204: Obtain target instruction information for indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the designated event on the target business.

[0050] In the embodiments of this specification, a large language model (LLM) may refer to a deep learning model with massive parameters obtained by pre-training based on massive text data. It can not only generate natural language text, but also deeply understand the meaning of the text and handle various natural language tasks such as text summarization, question answering, translation, etc. Since the large language model can be fine-tuned based on a small-scale data set in a specific field, it can produce good performance in tasks that require multi-hop reasoning capabilities (for example, common sense question answering and arithmetic reasoning) in a specific field, and thus is widely used to customize and solve complex analysis problems in fields such as retail, finance, and entertainment.

[0051] Since the large language model has the ability to understand natural language instructions to complete tasks, when the large language model is needed to perform the event impact assessment task, it is necessary to obtain target instruction information indicating the impact of the assessed event on the target business. By inputting the target instruction information into the large language model, the large language model can follow human instructions to perform the event impact assessment task.

[0052] In practical applications, large language models need to rely on semantic prior knowledge and learn from context to generate output information in a format that follows context examples. Specifically, large language models need to predict the next text output based on existing text (input text and historical generated text). Therefore, there must be a strong probabilistic association between its input and output. In order to use large language models to obtain the expected output, it is necessary to carefully design specific and clear inputs to increase their relevance to the expected output.

[0053] Based on this, when it is necessary to use a large language model to evaluate the impact of a target event on a target business, target instance information reflecting the historical impact of a specified event on the target business can be carried in the target instruction information, so that the large language model can learn knowledge from analogous target instance information, assist in understanding the event impact assessment task currently to be performed, and give an output result reflecting the impact of the target event on the target business based on task understanding.

[0054] In the embodiments of this specification, the target instance information is usually verified information with high credibility that reflects the historical impact of a specified event on a target business. Normally, one target instance information corresponds to one specified event, and the target instance information may usually include event information of the specified event and the historical impact of the specified event on the target business. In actual applications, since the amount of information in the event information of the specified event may be large, the summary information extracted from the event information of the specified event may also be carried in the target instance information, without the need to carry the full amount of event information in the target instance information, which is beneficial for saving data storage space.

[0055] In actual applications, service providers can usually generate a preset instance information set based on historical data and manual experience, and the preset instance information set can include multiple preset instance information, so as to select and configure the target instance information carried in the target instruction information from the preset instance information set. Among them, one preset instance information usually only corresponds to one event, and the preset instance information can include event information / event summary information of the event and the historical impact of the event on the target business, which will not be elaborated.

[0056] Step 206: Generate a target prompt word according to the target instruction information and the event information of the target event.

[0057] In the embodiments of this specification, prompt words (Prompt) may refer to artificial intelligence (AI) prompt words, which is a method of using natural language to guide or stimulate the NPC language model to complete specific tasks. Simply put, Prompt is a text prompt that can guide AI to complete specific tasks, such as generating text, translating, answering questions, etc. At present, the application scenarios of Prompt are also very rich. In the field of chatbots, Prompt can be used to generate more humane and intelligent conversations; in the field of search engines, Prompt can be used to optimize the quality of search results; in the field of intelligent customer service, Prompt can be used to improve the response speed and resolution rate of customer service, etc. Based on this, when it is necessary to use a large language model to evaluate the impact of a target event on a target business, it is necessary to generate a target prompt word containing the target instruction information and the event information of the target event.

[0058] Step 208: Using a large language model, based on the target prompt word, generate an evaluation result reflecting the impact of the target event on the target business.

[0059] In the embodiments of this specification, the target prompt word can be input into a large language model so that the large language model understands and analyzes the target prompt word to obtain an evaluation result that accurately reflects the impact of the target event on the target business. In practical applications, the impact reflected in the evaluation result can be implemented in a variety of ways. For example, the impact can be a few preset impact levels with a coarser granularity, or the impact can also be dozens or even hundreds of preset impact scores with a finer granularity. It is only necessary to allow the user to more accurately quantify the impact of the target event on the target business based on the evaluation result, and no further elaboration is given.

[0060] Figure 2 The method generates a target prompt word by combining event information of a target event related to a target business and target instruction information for indicating the impact of the evaluation event on the target business. In addition, the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business. Based on the super natural language understanding and natural language generation capabilities of the large language model, the large language model can better understand and infer the meaning of the input target prompt word, so as to conveniently generate an evaluation result that accurately reflects the impact of the target event on the target business, thereby realizing an accurate quantitative evaluation of the relationship and impact between the target event and the operation of the target business.

[0061] based on Figure 2 The method in this specification also provides some specific implementation plans of the method, which are described below.

[0062] In the embodiments of this specification, since the quality of the Prompt has a significant impact on the quality of the output information of the large language model, and high-quality Prompts often need to be supplemented with more details to enhance the understanding of the large language model, and also need to be simplified to clarify the requirements, in order to ensure that the output information of the large language model meets our actual expectations. Based on this, the service provider will usually pre-configure multiple preset instance information to form a preset instance information set, and will pre-generate multiple preset instruction information based on the preset instance information to form a preset instruction information set, so that when performing an event impact assessment, the target instruction information with better performance can be screened out from the preset instruction information set according to actual needs to ensure the excellent performance of the generated target prompt words.

[0063] Based on this, step 204: obtaining target instruction information for indicating the impact of the evaluation event on the target business may specifically include:

[0064] The similarity between the event information of the target event and the preset instance information in the preset instance information set is calculated; wherein, one of the preset instance information is used to reflect the historical impact of an event on the target business.

[0065] The preset instance information to which the maximum value of the similarity belongs is determined to obtain the target instance information.

[0066] The target instruction information is determined from the preset instruction information in the preset instruction information set that carries the target instance information.

[0067] In the embodiment of this specification, when the similarity between the preset instance information and the event information of the target event is large, if the preset instance information is used as the sample information in the Prompt, the output information generated by the large language model according to the Prompt is often more able to meet the user's impact assessment requirements for the target event and the target business. Based on this, the target instance information with the greatest similarity to the event information of the target event can be determined from the preset instance information set.

[0068] In practical applications, there are many ways to calculate the similarity between the event information of the target event and the preset instance information. For example, in order to reduce the amount of calculation and improve the accuracy of the similarity, the similarity between the event summary information of the target event and the event summary information in the preset instance information can be specifically calculated. Alternatively, a neural network model or a large language model can be used to extract corresponding event feature vectors from the event information of the target event and the event information in the preset instance information to calculate the similarity between the event feature vectors, which is not specifically limited.

[0069] In the embodiments of the present specification, since any preset instruction information in the preset instruction information set is often configured with several preset instance information in the preset instance information set, and the preset instance information configured in different preset instruction information is often not completely consistent. When it is necessary to generate a Prompt in combination with the target instance information, it is necessary to determine the preset instruction information carrying the target instance information from the preset instruction information set as the target instruction information. It is understandable that, in addition to the target instance information having the greatest similarity with the event information of the target event, the target instruction information may also include other preset instruction information pre-bound to the target instruction information, or may not include other preset instruction information, and there is no specific limitation on this.

[0070] The determining the target instruction information from the preset instruction information in the preset instruction information set that carries the target instance information may include:

[0071] Acquire a historical revenue value of the preset instruction information carrying the target instance information; wherein the historical revenue value is used to reflect the revenue generated by using the preset instruction information to evaluate the historical impact of an event on the target business.

[0072] The preset instruction information to which the maximum value of the historical profit value belongs is determined as the target instruction information.

[0073] In the embodiments of the present specification, the preset instruction information in the preset instruction information set is usually provided with a historical benefit value, wherein the higher the historical benefit value, the more accurate the impact of the event on the target business can be when the preset instruction information is used to evaluate the impact of the event on the target business, and the lower the historical benefit value, the less accurate the impact of the event on the target business can be when the preset instruction information is used to evaluate the impact of the event on the target business. Based on this, if there are multiple preset instruction information carrying the target instance information in the preset instruction information set, the preset instruction information carrying the target instance information with the highest historical benefit value can be used as the target instruction information.

[0074] In practical applications, the historical revenue value of the preset instruction information may be manually set, or may be determined based on the validity verification result and availability verification result of the corresponding impact evaluation result after evaluating the historical impact of historical events on the target business using the preset instruction information. Since the implementation principle of determining the (historical) revenue value of the corresponding instruction information based on the validity verification result and availability verification result of the impact evaluation result will be explained in detail in the subsequent embodiments, it will not be described in detail here.

[0075] For ease of understanding, the implementation principle of generating a target prompt word by combining target instruction information with event information of a target event is given here.

[0076] Specifically, the target instruction information may further include: task description information for instructing the large language model to evaluate the impact of the preset event information on the target business; and the task description information may further include business description information for the target business;

[0077] Correspondingly, step 206: generating a target prompt word according to the target instruction information and the event information of the target event may specifically include:

[0078] The event information of the target event is set as the preset event information carried in the task description information to obtain the target prompt word.

[0079] In the embodiment of this specification, the target instruction information may include two types of information, namely, task description information for instructing the large language model to evaluate the impact of preset event information on the target business, and instance information reflecting the historical impact of the specified event on the target business. The task description information may specifically include: business description information for the target business, task content introduction information, and preset event information waiting to be filled in.

[0080] For ease of understanding, an example of a target prompt word is given here. Specifically, the target prompt word can be expressed as:

[0081] [The target business is used to provide XX service to users, and the service period is XX years;] (i.e., business description information for the target business).

[0082] [The following is information about instances where impact has been assessed:

[0083] Specify the event (summary) information of event 1, and specify that the historical impact of event 1 on the target business is XX;

[0084] The event (summary) information of the specified event 2, the historical impact of the specified event 2 on the target business is XX. ……] (ie, the instance information carried in the target instruction information).

[0085] [Please evaluate the impact of the following events on the target business based on the above business description information and instance information:] (i.e. task content introduction information).

[0086] [1. Event (summary) information of target event 1;

[0087] 2. Event (summary) information of target event 2. […] (i.e., preset event information after filling).

[0088] It can be understood that the specific content in the above examples does not serve as a specific limitation on the target prompt words in the embodiments of this specification. In actual applications, the number of instance information and the number of target events in the target prompt words can also be other positive integers, and the business description information and task description information in the target prompt words can also be in other forms, and there is no specific limitation on this.

[0089] Since the instruction information used in the impact assessment process is closely related to the output quality of the large language model, there is often a need to continuously optimize the preset instruction information set.

[0090] Based on this, after step 208: generating an evaluation result reflecting the impact of the target event on the target business according to the target prompt word, the following may also be included:

[0091] The benefit value of the target instruction information is determined according to the validity verification result of the evaluation result; wherein the validity verification result is used to reflect the recommendation effect of using the event information of the target event as user recommended content.

[0092] The reinforcement learning training samples generated based on the event information of the target event, the target instruction information and the benefit value of the target instruction information are divided into a training sample set.

[0093] Based on the reinforcement learning algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; wherein the policy network model is used to receive event information samples, and output instruction information samples for evaluating the impact of the event information samples on the target business; the value network model is used to receive the event information samples and the instruction information samples, and output a value that reflects the benefits generated by using the instruction information samples to evaluate the impact of the event information samples on the target business.

[0094] The target strategy network model is used to optimize the preset instruction information set to which the target instruction information belongs, so as to obtain an optimized preset instruction information set.

[0095] In the embodiment of this specification, the service provider uses Figure 2After obtaining the evaluation result of the impact of the target event on the target business, the solution in the above method can often generate user recommendation content based on the event information of the target event, and push the user recommendation content to the user group by means such as text messages and application notification messages, so that the user group can process the target business according to the user recommendation content. When the user group gives good feedback on the user recommendation content, it can often mean that the user group is more in favor of the evaluation result of the impact of the target event on the target business. At this time, it can be said that the quality of the target instruction information used in the impact assessment is good. Based on this, the service provider can obtain the validity verification result of the evaluation result by sending recommended content to the user group, and then determine the benefit value of the target instruction information.

[0096] In the embodiments of this specification, reinforcement learning (RL) is one of the paradigms and methodologies of machine learning, which is used to describe and solve the problem of an agent (agent) learning a strategy to maximize rewards or achieve specific goals during the interaction with the environment. Reinforcement learning regards learning as a trial and evaluation process. The agent selects an action for the environment. The state of the environment changes after accepting the action, and a reinforcement signal (reward or punishment) is generated and fed back to the agent. The agent selects the next action based on the reinforcement signal and the current state of the environment. The principle of selection is to increase the probability of receiving positive reinforcement (reward). The selected action not only affects the immediate reinforcement value, but also affects the state of the environment at the next moment and the final reinforcement value. Since reinforcement learning can realize autonomous learning and decision-making of the agent, it has strong flexibility and adaptability, and is widely used in strategy optimization scenarios.

[0097] In practical applications, the reinforcement learning process usually involves two network models: the policy network model (Actor network model) and the value network model (Critic network model). The Actor network model is responsible for selecting appropriate actions based on the current state, while the Critic network model is responsible for evaluating the value function of the state and action. By combining these two parts for training, the Actor network model can gradually learn the optimization strategy and make better decisions in a changing environment.

[0098] Based on this, when it is necessary to optimize the preset instruction information required for the event impact process, it can be implemented based on the reinforcement learning algorithm. Specifically, the event information of the target event, the target instruction information, and the benefit value of the target instruction information can be used to generate reinforcement learning training samples. And train the strategy network model (Actor network) and the value network model (Critic network) to use the trained target strategy network model to generate optimized instruction information with better quality, and then use the optimized instruction information to optimize the preset instruction information set, which is convenient and fast.

[0099] In the embodiment of this specification, determining the benefit value of the target instruction information according to the validity verification result of the evaluation result may specifically include:

[0100] Obtaining preset feedback index values ​​of user groups for user recommendation content generated based on event information of the target event; wherein the preset feedback index values ​​include: at least one of adoption rate, reply rate and business conversion rate.

[0101] The benefit value of the target instruction information is determined according to the preset feedback indicator value.

[0102] In the embodiments of this specification, when generating user recommendation content based on the event information of the target event, the detailed event information / event summary information of the target event can be carried in the user recommendation content, or the evaluation result of the impact of the target event can be carried in the user recommendation content to improve the user experience. In addition, the business description information for the target business can also be carried in the user recommendation content. And, the processing suggestion information for the target business can also be carried in the user recommendation content, which is not specifically limited.

[0103] In actual applications, after viewing the user recommended content, if the user group believes that the user recommended content, they can perform an operation, which also means that the individual has adopted the user recommended content, or they can perform a reply operation on the user recommended content, or they can also perform a processing operation on the target business according to the user recommended content (for example, order, unsubscribe, collect, follow, etc.). The service provider can obtain various operations performed by the user group on the user recommended content, and thus obtain the preset feedback index value of the user recommended content generated by the user group for the event information based on the target event, which will not be elaborated.

[0104] In the embodiment of this specification, the benefit value of the target instruction information can be determined based on the preset feedback index value based on the preset preset strategy. The preset strategy can be set according to actual needs and is not specifically limited to this. However, under normal circumstances, the preset feedback index value and the benefit value of the target instruction information can be positively correlated.

[0105] When there are a large number of target events, it is usually necessary to prioritize target events with higher influence as user recommendation content in order to improve the accuracy and effectiveness of user recommendation content.

[0106] Based on this, before determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, the following may also be included:

[0107] According to the impact of each of the target events on the target business reflected by each of the evaluation results, each of the target events is sorted to obtain a target event sequence.

[0108] According to the ranking of each of the target events in the target event sequence, the priority of using the event information of each of the target events as the user recommendation content related to the target business is determined.

[0109] According to the priority corresponding to each of the target events, the user recommended content generated based on at least part of the event information of the target event is sent to the user device.

[0110] In the embodiments of this specification, since the higher the impact of the target event on the target business, it can usually mean that the target event has a higher impact on the user's decision to handle the target business, that is, the user has a higher interest in and attention to the target event, so the priority of each target event as a user recommendation content can be determined according to the order of the impact of each target event on the target business. Generally, the greater the impact of the target event on the target business, the higher the priority of the target event as a user recommendation content.

[0111] In actual applications, user recommendation content can be generated based on the event information of a preset number of target events with the highest priority, and the user recommendation content can be sent to the user devices used by the user group by means such as text messages, application notification information, emails, instant messaging information, etc. In addition, personalized information recommendation can be performed for each user, and user recommendation content can be generated for target events that the user may be interested in but have a lower priority based on the user's interest points or preference information, and the user recommendation content can be sent to the user device of the user in a targeted manner, without specific limitation.

[0112] Since it is usually necessary to send user recommendation content with high credibility and compliant content to the user group, when the credibility of the evaluation result of the impact of the target event is poor, or the event information of the target event is not compliant, it will affect the decision of whether to generate user recommendation content based on the event information of the target event, and further affect the benefit value of the target instruction information.

[0113] Based on this, determining the benefit value of the target instruction information may specifically include:

[0114] Determine a usability verification result for the evaluation result; wherein the usability verification result is used to reflect whether the event information of the target event can be recommended to the user as user recommendation content related to the target business.

[0115] The benefit value of the target instruction information is determined according to the usability verification result and the validity verification result for the evaluation result.

[0116] For ease of understanding, an implementation method for performing usability verification on the evaluation result is provided herein.

[0117] Specifically, determining the usability verification result for the evaluation result may include:

[0118] Determine whether there is preset instance information in the preset instance information set to which the target instance information belongs, whose similarity with the event information of the target event is greater than a first threshold, and / or determine whether there is preset instance information in the preset instance information set, whose similarity with the evaluation result corresponding to the target event is greater than a second threshold, and / or determine whether the evaluation result can pass the review of the security operation review strategy to obtain an availability verification result for the evaluation result.

[0119] In the embodiments of the present specification, since the preset instance information contained in the preset instance information set usually has good credibility, when there is preset instance information in the preset instance information set whose similarity with the event information of the target event is greater than a first threshold, it can be indicated that the event information of the target event has good credibility, and thus an availability verification result indicating that the availability of the evaluation result is good can be generated.

[0120] Since both the evaluation result and the preset instance information may include event information and event impact information, in addition to performing a similarity comparison between the evaluation result and the event information contained in the preset instance information, a similarity comparison may also be performed between the event impact information in the evaluation result and the historical impact information of the events carried in the preset instance information on the target business. When the similarity of the two types of information is greater than a threshold, an availability verification result indicating that the availability of the evaluation result is better is generated. There is no specific limitation on this.

[0121] In addition, when there is no event information with the target event or the preset instance information whose similarity with the evaluation result is greater than a threshold, manual or security operation audit strategies can be used to identify whether the evaluation result is credible and whether the event information of the target event is compliant. If so, an availability verification result indicating that the evaluation result has better availability can be generated.

[0122] In practical applications, corresponding benefit values ​​may be pre-set for each type of usability verification result, so as to determine the first benefit value of the target instruction information based on the usability verification result. In addition, the second benefit value of the target instruction information may also be determined based on the validity verification result. The sum of the first benefit value and the second benefit value is calculated to obtain the benefit value of the target instruction information.

[0123] For example, when there is preset instance information whose similarity with the event information of the target event or the evaluation result is greater than a threshold, the target instruction information can be assigned a third benefit value; and when the evaluation result is identified as credible using the security operation audit strategy, the target instruction information is assigned a fourth benefit value, and when the evaluation result is identified as unreliable using the security operation audit strategy, the target instruction information is assigned a fifth benefit value; and, when the event information of the target event is identified as compliant using the security operation audit strategy, the target instruction information is assigned a sixth benefit value, otherwise, the target instruction information is assigned a seventh benefit value; according to actual conditions, at least part of the third to seventh benefit values ​​are summed up to obtain the first benefit value.

[0124] In addition, when the validity verification results include adoption rate, response rate and business conversion rate, the target instruction information can also be assigned an eighth benefit value, a ninth benefit value and a tenth benefit value according to each preset feedback indicator value, and the sum of the eighth benefit value, the ninth benefit value and the tenth benefit value can be calculated to obtain the second benefit value, which will not be elaborated on.

[0125] In the embodiment of the specification, the reinforcement learning training samples generated based on the event information of the target event, the target instruction information, and the benefit value of the target instruction information are divided into a training sample set, which may specifically include:

[0126] Obtain summary information of each event topic in the event information of the target event.

[0127] The summary information of any of the event topics is set as the state value at the target moment, the target instruction information is set as the action value at the target moment, the benefit value of the target instruction information is set as the benefit value at the target moment, and the summary information of another event topic is set as the state value at the next moment after the target moment, to obtain a reinforcement learning training sample.

[0128] In the embodiment of the present specification, since the amount of information contained in the event information of the target event obtained in step 202 is usually relatively rich, the event information of the target event can often contain relevant information of multiple event topics. In order to reduce the amount of data of the reinforcement learning training sample and improve the efficiency of model training, the summary information of each event topic can be first extracted from the event information of the target event. In addition, the summary information of two consecutive event topics can be set as the state value (St) at the target moment and the state value (St+1) at the next moment, respectively, and the target instruction information used for the impact assessment can be used as the action value (At) at the target moment, and the benefit value of the target instruction information is set to the benefit value (Rt) at the target moment. It can be seen that when the event information of the target event contains more than two event topics, multiple reinforcement learning training samples can be generated based on the event information of the target event, so that the data can be fully utilized.

[0129] In the embodiment of this specification, the reinforcement learning algorithm is used to train the policy network model and the value network model using the training sample set to obtain the target policy network model and the target value network model, which may specifically include:

[0130] Based on the deep deterministic policy gradient algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or,

[0131] Based on the trust region strategy optimization algorithm, the strategy network model and the value network model are trained using the training sample set to obtain a target strategy network model and a target value network model; or,

[0132] Based on the proximal policy optimization algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or,

[0133] Based on the actor-critic algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model.

[0134] In the embodiments of this specification, the deep deterministic policy gradient algorithm, the trust region policy optimization algorithm, the proximal policy optimization algorithm, the actor-critic algorithm, etc. are all relatively classic reinforcement learning algorithms, and thus the specific process of obtaining the target policy network model and the target value network model by training based on the above algorithms using training samples will not be described in detail.

[0135] In the embodiment of this specification, the use of the target policy network model to optimize the preset instruction information set to which the target instruction information belongs may specifically include:

[0136] The event information of the preset event is input into the target strategy network model to obtain the optimization instruction information output by the target strategy network model.

[0137] Add the optimization instruction information to the preset instruction information set; or,

[0138] Obtain optimization instance information from the optimization instruction information.

[0139] adding at least part of the optimization instance information to the preset instruction information in the preset instruction information set; or,

[0140] At least part of the optimized instance information is used to replace part of the instance information carried in the preset instruction information.

[0141] In the embodiments of the present specification, since the target policy network model can output optimization instruction information that can be used to evaluate the impact of the event information based on the received event information, and the optimization instruction information can include optimization task description information and optimization instance information, the preset instruction information set can be optimized using the optimization instruction information. Among them, the optimization task description information can include optimized information for indicating the impact of the large language model-assessed event on the target business, and can also include business description information for the target business. The optimization instance information can include event information that has occurred or has not occurred that performs well when paired with the optimization task description information. In addition, the optimization instance information can also include information reflecting the historical impact of the corresponding event information on the target business, and there is no specific limitation on this.

[0142] Specifically, the optimization instruction information can be directly added as newly added preset instruction information to the preset instruction information set. Alternatively, part or all of the optimization instance information contained in the optimization instruction information can be added to one or several original preset instruction information to increase the richness of the instance information in the preset instruction information. Alternatively, part or all of the optimization instance information contained in the optimization instruction information can be used to replace part or all of the instance information in one or several original preset instruction information to update the original preset instruction information. Good flexibility.

[0143] In actual applications, when screening preset instruction information that needs to be optimized, it can be implemented based on manual experience, or the preset instruction information whose similarity with the optimization task description information and / or at least part of the optimization instance information in the optimization instruction information is greater than a threshold can be used as the preset instruction information that needs to be optimized and updated according to the optimization instruction information to ensure the optimization effect.

[0144] In the present embodiment of the specification, due to Figure 2 The target instance information used in the middle scheme for event impact assessment is usually stored in a preset instruction information set, and is pre-configured with historical benefit values ​​that affect the usage frequency of the target instance information. Therefore, after the benefit value of the latest version of the target instruction information is calculated this time, the historical benefit value of the target instruction information can be updated to facilitate subsequent decisions based on the historical benefit value of the latest version of the target instruction information whether to continue to use it to assess the impact of the event, which is conducive to improving the accuracy of the impact assessment results.

[0145] Based on this, after determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, the method may further include:

[0146] The historical profit value of the target instruction information in the preset instruction information set is updated by using the profit value of the target instruction information to obtain the updated historical profit value of the target instruction information.

[0147] In the embodiments of the present specification, the historical benefit values ​​of the preset instruction information in the preset instruction information set can also be periodically updated based on manual experience, or based on the similarity between the instance information carried by each preset instruction information and the historical benefit values ​​of each preset instruction information, which will not be elaborated herein.

[0148] In the embodiment of this specification, after determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, the following may also be included:

[0149] It is determined whether the benefit value of the target instruction information reaches a third threshold value to obtain a determination result.

[0150] If the judgment result indicates that the benefit value of the target instruction information reaches the third threshold, new instance information is generated according to the event information of the target event and the evaluation result reflecting the impact of the target event on the target business.

[0151] The newly added instance information is added to the preset instance information set to which the target instance information belongs.

[0152] In the embodiment of this specification, if the benefit value of the target instruction information is high, it can be said that the event information of the target event is credible and compliant, and the impact assessment result of the target event generated in step 208 also has good credibility. Based on this, new instance information carrying the event (summary) information of the target event and the assessment result reflecting the impact of the target event on the target business can be generated and added to the preset instance information set, thereby continuously enriching the number of preset instance information in the preset instance information set. In addition, the preset instruction information in the preset instruction information set can also be generated or updated based on the newly added instance information, which will not be elaborated on.

[0153] Figure 3 The embodiments of this specification provide corresponding to Figure 2 A swimlane flow diagram of the event impact assessment method in . Figure 3 As shown, the event impact assessment process may involve execution entities such as service provider equipment and user equipment.

[0154] In the impact determination stage, the service provider's equipment can obtain event information of a target event related to the target business; determine the target instance information with the greatest similarity to the event information of the target event from the preset instance information set; determine the preset instruction information in the preset instruction information set that carries the target instance information and has the largest historical benefit value as the target instruction information; and set the event information of the target event to the preset event information carried by the task description information in the target instance information to obtain a target prompt word, thereby using a large language model to generate an evaluation result reflecting the impact of the target event on the target business based on the target prompt word.

[0155] In the optimization stage, the device of the service provider can generate and send user recommendation content generated based on at least part of the event information of the target event to the user device according to the impact of each target event on the target business reflected by the evaluation result. The user device can receive and display the user recommendation content to the user, so that the user can perform feedback operations on the user recommendation content. The device of the service provider can determine the benefit value of the target instruction information based on the preset feedback index value of the user group for the user recommendation content and the availability verification result reflecting whether the user recommendation content can be sent to the user.

[0156] Subsequently, the service provider's equipment may also divide the reinforcement learning training samples generated based on the event information of the target event, the target instruction information, and the benefit value of the target instruction information into a training sample set; based on the reinforcement learning algorithm, the policy network model and the value network model are trained using the training sample set to obtain the target policy network model and the target value network model. By inputting the event information of the preset event into the target policy network model, the optimization instruction information output by the target policy network model is obtained, so as to add the optimization instruction information to the preset instruction information set, and / or, add at least part of the optimization instance information in the optimization instruction information to the preset instruction information in the preset instruction information set, and / or, use at least part of the optimization instance information in the optimization instruction information to replace part of the instance information carried in the preset instruction information, thereby achieving optimization for the preset instruction information set.

[0157] In addition, the service provider's equipment can also use the benefit value of the target instruction information to update the historical benefit value of the target instruction information in the preset instruction information set. If the benefit value of the target instruction information reaches a third threshold, the event information based on the target event and the newly added instance information generated by the evaluation result can also be added to the preset instance information set, thereby continuously improving the accuracy of the event impact evaluation results.

[0158] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method. Figure 4 The embodiments of this specification provide corresponding to Figure 2 A schematic diagram of the structure of an event impact assessment device. Figure 4 As shown, the device may include:

[0159] The first acquisition module 402 is used to acquire event information of a target event related to a target business.

[0160] The second acquisition module 404 is used to acquire target instruction information indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the designated event on the target business.

[0161] The first generating module 406 is used to generate a target prompt word according to the target instruction information and the event information of the target event.

[0162] The second generating module 408 is used to generate an evaluation result reflecting the impact of the target event on the target business according to the target prompt word by using a large language model.

[0163] based on Figure 4 The present specification also provides some specific implementation schemes of the device, which are described below.

[0164] Optionally, the second acquisition module may include:

[0165] A calculation unit is used to calculate the similarity between the event information of the target event and the preset instance information in the preset instance information set; wherein, one of the preset instance information is used to reflect the historical impact of an event on the target business.

[0166] The first determining unit is used to determine the preset instance information to which the maximum value of the similarity belongs, and obtain the target instance information.

[0167] The second determining unit is used to determine the target instruction information from the preset instruction information carrying the target instance information in the preset instruction information set.

[0168] The second determining unit may be specifically configured to:

[0169] Acquire the historical revenue value of the preset instruction information carrying the target instance information; wherein the historical revenue value is used to reflect the revenue generated by using the preset instruction information to evaluate the historical impact of the event on the target business.

[0170] The preset instruction information to which the maximum value of the historical profit value belongs is determined as the target instruction information.

[0171] Optionally, the target instruction information may also include: task description information for instructing the large language model to evaluate the impact of preset event information on the target business; the task description information also includes business description information for the target business. Correspondingly, the first generation module may be specifically used to set the event information of the target event to the preset event information carried in the task description information to obtain the target prompt word.

[0172] Optional, Figure 4 The device described in the above method may further include:

[0173] The benefit value determination module is used to determine the benefit value of the target instruction information according to the validity verification result of the evaluation result; wherein the validity verification result is used to reflect the recommendation effect of using the event information of the target event as the user recommended content.

[0174] The sample generation module is used to divide the reinforcement learning training samples generated based on the event information of the target event, the target instruction information and the benefit value of the target instruction information into a training sample set.

[0175] A model training module is used to train the policy network model and the value network model based on the reinforcement learning algorithm using the training sample set to obtain a target policy network model and a target value network model; wherein the policy network model is used to receive event information samples and output instruction information samples for evaluating the impact of the event information samples on the target business; the value network model is used to receive the event information samples and the instruction information samples, and output a benefit that reflects the use of the instruction information samples to evaluate the impact of the event information samples on the target business.

[0176] The optimization module is used to optimize the preset instruction information set to which the target instruction information belongs by using the target strategy network model to obtain an optimized preset instruction information set.

[0177] Optionally, the revenue value determination module may be specifically used for:

[0178] Obtaining preset feedback index values ​​of user groups for user recommendation content generated based on event information of the target event; wherein the preset feedback index values ​​include: at least one of adoption rate, reply rate and business conversion rate.

[0179] The benefit value of the target instruction information is determined according to the preset feedback indicator value.

[0180] Optional, Figure 4 The device described in may also include:

[0181] A sorting module, used to sort each of the target events according to the impact of each of the target events on the target business reflected by each of the evaluation results, to obtain a target event sequence;

[0182] A priority determination module, used to determine the priority of using the event information of each target event as user recommendation content related to the target business according to the order of each target event in the target event sequence;

[0183] A sending module is used to send the user recommendation content generated based on at least part of the event information of the target event to the user device according to the priority corresponding to each of the target events.

[0184] Optionally, the revenue value determination module may also be used to:

[0185] Determine a usability verification result for the evaluation result; wherein the usability verification result is used to reflect whether the event information of the target event can be recommended to the user as user recommendation content related to the target business.

[0186] The benefit value of the target instruction information is determined according to the usability verification result and the validity verification result for the evaluation result.

[0187] The determining of the usability verification result for the evaluation result may specifically include:

[0188] Determine whether there is preset instance information in the preset instance information set to which the target instance information belongs, whose similarity with the event information of the target event is greater than a first threshold, and / or determine whether there is preset instance information in the preset instance information set, whose similarity with the evaluation result corresponding to the target event is greater than a second threshold, and / or determine whether the evaluation result can pass the review of the security operation review strategy to obtain an availability verification result for the evaluation result.

[0189] Optionally, the sample generation module may include:

[0190] The summary information acquisition unit is used to acquire summary information of each event theme in the event information of the target event.

[0191] A sample data setting unit is used to set the summary information of any of the event topics as the state value at the target moment, set the target instruction information as the action value at the target moment, set the benefit value of the target instruction information as the benefit value at the target moment, and set the summary information of another event topic as the state value at the next moment after the target moment, so as to obtain a reinforcement learning training sample.

[0192] The model training module can be specifically used for:

[0193] Based on the deep deterministic policy gradient algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or,

[0194] Based on the trust region strategy optimization algorithm, the strategy network model and the value network model are trained using the training sample set to obtain a target strategy network model and a target value network model; or,

[0195] Based on the proximal policy optimization algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or,

[0196] Based on the actor-critic algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model.

[0197] Optionally, the optimization module may include:

[0198] The first acquisition unit is used to input event information of a preset event into the target strategy network model to obtain optimization instruction information output by the target strategy network model.

[0199] A first adding unit is used to add the optimization instruction information to the preset instruction information set; or,

[0200] The second acquisition unit is used to acquire the optimization instance information in the optimization instruction information.

[0201] A second adding unit is configured to add at least part of the optimization instance information to the preset instruction information in the preset instruction information set; or

[0202] A replacement unit is used to replace part of the instance information carried in the preset instruction information with at least part of the optimization instance information.

[0203] Optional, Figure 4 The device described in the above method may further include:

[0204] An updating module is used to update the historical profit value of the target instruction information in the preset instruction information set by using the profit value of the target instruction information to obtain the updated historical profit value of the target instruction information.

[0205] Optional, Figure 4 The device described in the above method may further include:

[0206] A judgment module, used to judge whether the benefit value of the target instruction information reaches a third threshold value, and obtain a judgment result;

[0207] A third generating module is used to generate new instance information according to the event information of the target event and the evaluation result reflecting the impact of the target event on the target business if the judgment result indicates that the benefit value of the target instruction information reaches a third threshold;

[0208] The adding module is used to add the newly added instance information to the preset instance information set to which the target instance information belongs.

[0209] Based on the same idea, the embodiments of this specification also provide a device corresponding to the above method.

[0210] Figure 5 The embodiments of this specification provide corresponding to Figure 2 A schematic diagram of the structure of an event impact assessment device. Figure 5 As shown, the device 500 may include:

[0211] at least one processor 510; and,

[0212] A memory 530 in communication with the at least one processor; wherein,

[0213] The memory 530 stores instructions 520 executable by the at least one processor 510. The instructions are executed by the at least one processor 510 to enable the at least one processor 510 to:

[0214] Get event information of a target event related to a target business.

[0215] Target instruction information for indicating the impact of an evaluation event on the target business is obtained; wherein the target instruction information carries target instance information reflecting the historical impact of a designated event on the target business.

[0216] A target prompt word is generated according to the target instruction information and the event information of the target event.

[0217] By using a large language model, an evaluation result reflecting the impact of the target event on the target business is generated according to the target prompt word.

[0218] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. Figure 5 As for the device shown, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0219] In the 1990s, improvements to a technology could be clearly distinguished as hardware improvements (for example, improvements to the circuit structure of diodes, transistors, switches, etc.) or software improvements (improvements to the method flow). However, with the development of technology, many improvements to the method flow today can be regarded as direct improvements to the hardware circuit structure. Designers almost always obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement in a method flow cannot be implemented using a hardware entity module. For example, a programmable logic device (PLD) (such as a field programmable gate array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. Designers can "integrate" a digital system on a PLD by programming it themselves, without having to ask a chip manufacturer to design and produce a dedicated integrated circuit chip. Moreover, nowadays, instead of manually making integrated circuit chips, this kind of programming is mostly implemented by "logic compiler" software, which is similar to the software compiler used when developing and writing programs, and the original code before compilation must also be written in a specific programming language, which is called hardware description language (HDL). There is not only one HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones are VHDL (Very-High-Speed ​​Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also know that it is only necessary to program the method flow slightly in the above-mentioned hardware description languages ​​and program it into the integrated circuit, and then it is easy to obtain the hardware circuit that implements the logic method flow.

[0220] The controller can be implemented in any appropriate manner, for example, the controller can take the form of a microprocessor or processor and a computer-readable medium storing a computer-readable program code (such as software or firmware) that can be executed by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of controllers include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in a purely computer-readable program code manner, the controller can be implemented in the form of a logic gate, a switch, an application-specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, this controller can be considered as a hardware component, and the devices included therein for implementing various functions can also be regarded as structures within the hardware component. Or even, the devices for implementing various functions can be regarded as both software modules for implementing the method and structures within the hardware component.

[0221] The systems, devices, modules or units described in the above embodiments may be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0222] For the convenience of description, the above device is described in various units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0223] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0225] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0226] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0227] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0228] The memory may include non-permanent storage in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0229] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0230] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0231] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0232] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0233] The above is only an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of the claims of the present application.

Claims

1. A method for evaluating the impact of an event, comprising: Obtaining event information of a target event related to a target business; Acquire target instruction information for indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business; generating a target prompt word according to the target instruction information and the event information of the target event; By using a large language model, an evaluation result reflecting the impact of the target event on the target business is generated according to the target prompt word.

2. The method according to claim 1, wherein obtaining target instruction information for indicating the impact of the evaluation event on the target business specifically comprises: Calculating the similarity between the event information of the target event and the preset instance information in the preset instance information set; wherein one of the preset instance information is used to reflect the historical impact of an event on the target business; Determine the preset instance information to which the maximum value of the similarity belongs, and obtain the target instance information; The target instruction information is determined from the preset instruction information in the preset instruction information set that carries the target instance information.

3. The method according to claim 2, wherein determining the target instruction information from the preset instruction information in the preset instruction information set that carries the target instance information specifically comprises: Acquire a historical revenue value of the preset instruction information carrying the target instance information; wherein the historical revenue value is used to reflect the revenue generated by using the preset instruction information to evaluate the historical impact of an event on the target business; The preset instruction information to which the maximum value of the historical profit value belongs is determined as the target instruction information.

4. The method according to claim 1, wherein the target instruction information further includes: Task description information for instructing the large language model to evaluate the impact of preset event information on the target business; The generating of the target prompt word according to the target instruction information and the event information of the target event specifically includes: The event information of the target event is set as the preset event information carried in the task description information to obtain the target prompt word.

5. The method according to claim 4, wherein the task description information further includes business description information for the target business.

6. The method according to claim 1, after using the large language model to generate an evaluation result reflecting the impact of the target event on the target business according to the target prompt word, further comprising: Determine the benefit value of the target instruction information according to the validity verification result of the evaluation result; wherein the validity verification result is used to reflect the recommendation effect of using the event information of the target event as the user recommended content; Dividing the reinforcement learning training samples generated based on the event information of the target event, the target instruction information, and the benefit value of the target instruction information into a training sample set; Based on the reinforcement learning algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; wherein the policy network model is used to receive event information samples and output instruction information samples for evaluating the impact of the event information samples on the target business; the value network model is used to receive the event information samples and the instruction information samples, and output a benefit that reflects the use of the instruction information samples to evaluate the impact of the event information samples on the target business; The target strategy network model is used to optimize the preset instruction information set to which the target instruction information belongs, so as to obtain an optimized preset instruction information set.

7. The method according to claim 6, wherein determining the benefit value of the target instruction information according to the validity verification result of the evaluation result specifically comprises: Obtaining a preset feedback index value of a user group for user recommendation content generated based on the event information of the target event; wherein the preset feedback index value includes: at least one of an adoption rate, a reply rate, and a business conversion rate; The benefit value of the target instruction information is determined according to the preset feedback indicator value.

8. The method according to claim 6, before determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, further comprising: According to the impact of each of the target events on the target business reflected by each of the evaluation results, sorting each of the target events to obtain a target event sequence; Determining, according to the order of each of the target events in the target event sequence, the priority of using the event information of each of the target events as user recommendation content related to the target business; According to the priority corresponding to each of the target events, the user recommended content generated based on at least part of the event information of the target event is sent to the user device.

9. The method according to claim 6, wherein determining the benefit value of the target instruction information comprises: Determine a usability verification result for the evaluation result; wherein the usability verification result is used to reflect whether the event information of the target event can be recommended to the user as user recommendation content related to the target business; The benefit value of the target instruction information is determined according to the usability verification result and the validity verification result for the evaluation result.

10. The method according to claim 9, wherein determining a usability verification result for the evaluation result comprises: Determine whether there is preset instance information in the preset instance information set to which the target instance information belongs, whose similarity with the event information of the target event is greater than a first threshold, and / or determine whether there is preset instance information in the preset instance information set, whose similarity with the evaluation result corresponding to the target event is greater than a second threshold, and / or determine whether the evaluation result can pass the review of the security operation review strategy to obtain an availability verification result for the evaluation result.

11. The method according to claim 6, wherein the reinforcement learning training samples generated based on the event information of the target event, the target instruction information, and the benefit value of the target instruction information are divided into a training sample set, specifically comprising: Obtain summary information of each event topic in the event information of the target event; The summary information of any of the event topics is set as the state value at the target moment, the target instruction information is set as the action value at the target moment, the benefit value of the target instruction information is set as the benefit value at the target moment, and the summary information of another event topic is set as the state value at the next moment after the target moment, to obtain a reinforcement learning training sample.

12. The method according to claim 11, wherein the method is based on a reinforcement learning algorithm and uses the training sample set to train the policy network model and the value network model to obtain a target policy network model and a target value network model, specifically comprising: Based on the deep deterministic policy gradient algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or, Based on the trust region strategy optimization algorithm, the strategy network model and the value network model are trained using the training sample set to obtain a target strategy network model and a target value network model; or, Based on the proximal policy optimization algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model; or, Based on the actor-critic algorithm, the policy network model and the value network model are trained using the training sample set to obtain a target policy network model and a target value network model.

13. The method according to claim 6, wherein the step of optimizing the preset instruction information set to which the target instruction information belongs by using the target policy network model comprises: Inputting event information of a preset event into the target strategy network model to obtain optimization instruction information output by the target strategy network model; Add the optimization instruction information to the preset instruction information set; or, Obtaining optimization instance information in the optimization instruction information; adding at least part of the optimization instance information to the preset instruction information in the preset instruction information set; or, At least part of the optimized instance information is used to replace part of the instance information carried in the preset instruction information.

14. The method according to claim 6, after determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, further comprising: The historical profit value of the target instruction information in the preset instruction information set is updated by using the profit value of the target instruction information to obtain the updated historical profit value of the target instruction information.

15. The method according to claim 6, after determining the benefit value of the target instruction information according to the validity verification result of the evaluation result, further comprising: Determine whether the benefit value of the target instruction information reaches a third threshold value, and obtain a determination result; If the judgment result indicates that the benefit value of the target instruction information reaches a third threshold, then generating new instance information according to the event information of the target event and the evaluation result reflecting the impact of the target event on the target business; The newly added instance information is added to the preset instance information set to which the target instance information belongs.

16. An event impact assessment device, comprising: A first acquisition module, used to acquire event information of a target event related to a target business; A second acquisition module is used to acquire target instruction information indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business; A first generating module, configured to generate a target prompt word according to the target instruction information and the event information of the target event; The second generating module is used to generate an evaluation result reflecting the impact of the target event on the target business according to the target prompt word by using a large language model.

17. The device according to claim 16, wherein the second acquisition module comprises: A calculation unit, used to calculate the similarity between the event information of the target event and the preset instance information in the preset instance information set; wherein one of the preset instance information is used to reflect the historical impact of an event on the target business; A first determining unit, configured to determine the preset instance information to which the maximum value of the similarity belongs, and obtain the target instance information; A second determining unit, configured to determine the target instruction information from the preset instruction information carrying the target instance information in the preset instruction information set; Wherein, the second determining unit is specifically used to: Acquire a historical revenue value of the preset instruction information carrying the target instance information; wherein the historical revenue value is used to reflect the revenue generated by using the preset instruction information to evaluate the historical impact of an event on the target business; The preset instruction information to which the maximum value of the historical profit value belongs is determined as the target instruction information.

18. The device according to claim 16, wherein the target instruction information further includes: Task description information for instructing the large language model to evaluate the impact of preset event information on the target business; the task description information also includes business description information for the target business; The first generating module is specifically configured to set the event information of the target event as the preset event information carried in the task description information to obtain the target prompt word.

19. The apparatus of claim 16, further comprising: A benefit value determination module, used to determine the benefit value of the target instruction information according to the validity verification result of the evaluation result; wherein the validity verification result is used to reflect the recommendation effect of using the event information of the target event as the user recommended content; A sample generation module, used for dividing the reinforcement learning training samples generated based on the event information of the target event, the target instruction information and the benefit value of the target instruction information into a training sample set; A model training module, used to train the policy network model and the value network model using the training sample set based on a reinforcement learning algorithm to obtain a target policy network model and a target value network model; wherein the policy network model is used to receive event information samples and output instruction information samples for evaluating the impact of the event information samples on the target business; the value network model is used to receive the event information samples and the instruction information samples, and output a benefit reflecting the use of the instruction information samples to evaluate the impact of the event information samples on the target business; The optimization module is used to optimize the preset instruction information set to which the target instruction information belongs by using the target strategy network model to obtain an optimized preset instruction information set.

20. The device according to claim 19, wherein the revenue value determination module is further configured to: Determine a usability verification result for the evaluation result; wherein, The availability verification result is used to reflect whether the event information of the target event can be recommended to the user as user recommendation content related to the target business; The benefit value of the target instruction information is determined according to the usability verification result and the validity verification result for the evaluation result.

21. The apparatus according to claim 19, wherein the sample generation module comprises: A summary information acquisition unit, used to acquire summary information of each event theme in the event information of the target event; A sample data setting unit is used to set the summary information of any of the event topics as the state value at the target moment, set the target instruction information as the action value at the target moment, set the benefit value of the target instruction information as the benefit value at the target moment, and set the summary information of another event topic as the state value at the next moment after the target moment, so as to obtain a reinforcement learning training sample.

22. The apparatus of claim 19, wherein the optimization module comprises: A first acquisition unit, configured to input event information of a preset event into the target strategy network model to obtain optimization instruction information output by the target strategy network model; A first adding unit, configured to add the optimization instruction information to the preset instruction information set; or, A second acquisition unit, used to acquire optimization instance information in the optimization instruction information; A second adding unit is configured to add at least part of the optimization instance information to the preset instruction information in the preset instruction information set; or A replacement unit is used to replace part of the instance information carried in the preset instruction information with at least part of the optimization instance information.

23. The apparatus of claim 19, further comprising: An updating module is used to update the historical profit value of the target instruction information in the preset instruction information set by using the profit value of the target instruction information to obtain the updated historical profit value of the target instruction information.

24. The apparatus of claim 19, further comprising: A judgment module, used to judge whether the benefit value of the target instruction information reaches a third threshold value, and obtain a judgment result; A third generating module is used to generate new instance information according to the event information of the target event and the evaluation result reflecting the impact of the target event on the target business if the judgment result indicates that the benefit value of the target instruction information reaches a third threshold; The adding module is used to add the newly added instance information to the preset instance information set to which the target instance information belongs.

25. An event impact assessment device, comprising: at least one processor; as well as, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to: Obtaining event information of a target event related to a target business; Acquire target instruction information for indicating the impact of the evaluation event on the target business; wherein the target instruction information carries target instance information reflecting the historical impact of the specified event on the target business; generating a target prompt word according to the target instruction information and the event information of the target event; By using a large language model, an evaluation result reflecting the impact of the target event on the target business is generated according to the target prompt word.

Images