Information service method, device and equipment, storage medium and program product

Abstract

The invention provides an information service method, device and equipment, a storage medium and a program product, and relates to the field of artificial intelligence. The information service method comprises the steps that in response to the fact that a first interface is triggered, information of an event recommended for a user is displayed on the first interface, and the information of the event comprises summary information of an exclusive solution recommended for the user and at least part of recommendation basis information of the solution. Therefore, personalized information service in an event form is actively provided for the user, different from traditional static content in an object form, the information in the event form comprises summary information of a solution customized for the user and at least part of recommendation basis information of the solution. Therefore, the user feels that the user is understood and enjoys the exclusive service, the interaction desire of the user is stimulated, and the reliability and the use stickiness of the user are improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the field of artificial intelligence, and particularly relates to an information service method, device and equipment, a storage medium and a program product. BACKGROUND

[0002] With the improvement of big data and computing power, artificial intelligence technology has made significant progress, especially in model inference ability and multi-modal fusion, which has promoted the development and wide application of intelligent interaction.

[0003] There are two mainstream modes of user-oriented information services based on artificial intelligence technology.

[0004] (1) Instruction response mode: the user initiates an explicit instruction, such as a search instruction, to the artificial intelligence system, and the artificial intelligence system responds to the user's instruction and feeds back the processing result of the user's instruction.

[0005] This mode requires the user to clearly understand their own needs and organize them into explicit instructions, which has a high requirement for the user's demand expression ability. When the user's own needs are vague and cannot accurately describe the instruction, the artificial intelligence system is difficult to output effective services.

[0006] (2) Traditional content recommendation mode: the user passively receives the push content of the artificial intelligence system, and the push content is a static "thing" form, such as a specific product or a specific video.

[0007] The content recommended by this mode is relatively rigid and difficult to stimulate the user's interaction desire, which affects the user's trust and use stickiness of the artificial intelligence system. SUMMARY

[0008] The information service scheme of the embodiments of the present disclosure can actively provide personalized "event" form information services for users, which is different from the traditional static "thing" form content. The "event" form information includes abstract information of a "tailor-made" solution for the user and at least part of the recommendation basis information of the solution, so that the user feels that they are understood and enjoy exclusive services, stimulating the user's interaction desire and improving the user's trust and use stickiness.

[0009] Some embodiments of the present disclosure provide an information service method, comprising: in response to a first interface being triggered, displaying information of an event recommended for a user in the first interface, the information of the event including abstract information of an exclusive solution recommended for the user and at least part of the recommendation basis information of the solution.

[0010] In some embodiments, the recommendation basis information of the solution includes one or more of the user's state, the user's potential needs, environmental information, and incentives.

[0011] In some embodiments, the event is associated with an entry path of the intelligent agent, and the method further comprises: in response to the triggering operation of the user on the event, displaying an execution result of the solution from the intelligent agent on the second interface.

[0012] In some embodiments, the execution result of the solution comprises a recommended item, and the recommended item comprises a recommended reason, which is determined according to the solution of the user and the recommendation basis information.

[0013] In some embodiments, the execution result of the solution is obtained by the intelligent agent according to the recommendation basis information of the solution.

[0014] In some embodiments, the information of the event comprises a reference picture of the solution, which is from a picture of a product related to the solution or generated according to the solution.

[0015] In some embodiments, the type of the event comprises one or more of a shopping type, a takeout type, an order type, an article type, a special topic type, a love purchase type, a repeat purchase type, a price reduction type, a combination purchase type, a usage guide type, and a question type.

[0016] In some embodiments, the reference picture of the solution of the event of the order type, the love purchase type, the repeat purchase type, the price reduction type, or the usage guide type is from a picture of a product related to the solution, or the reference picture of the solution of the event of the article type, the special topic type, the combination purchase type, or the question type is generated according to the solution, or the reference picture of the solution of the event of the shopping type or the takeout type is from a picture of a product related to the solution or generated according to the solution.

[0017] In some embodiments, the method further comprises:

[0018] Based on the thought chain capability, performing a prior abductive reasoning according to the behavior information of the user and the memory information for the user to obtain a candidate event set of the user;

[0019] In response to the first interface being triggered, performing a generative sorting on the candidate events in the candidate event set of the user to obtain an event recommended for the user;

[0020] Displaying the information of the event on the first interface.

[0021] In some embodiments, the memory information for the user comprises a memory slice for the user and a context logic formed by the memory slice.

[0022] In some embodiments, the memory information for the user includes: a preference of the user in a first period, a state of the user in a second period, and an intention of the user in a third period, wherein the first period is greater than the second period, and the second period is greater than the third period.

[0023] In some embodiments, the abductive reasoning is performed by using an event reasoning model, and an updating method of the event reasoning model includes:

[0024] For a group of candidate events in the set of candidate events generated based on the same behavior information and memory information of the user, a reward of each candidate event in the group is calculated;

[0025] According to the reward of each candidate event in the group, a relative advantage of each candidate event in the group is calculated.

[0026] According to the relative advantages of all candidate events in the group, an inference strategy of the event reasoning model is updated.

[0027] In some embodiments, calculating the reward of each candidate event in the group includes: calculating the reward of each candidate event in the group according to at least one of logical self-consistency, surprise degree, format, interaction rate, task completion rate, retention value, and disturbance penalty value of each candidate event in the group, wherein the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the candidate event are determined according to the feedback information of the user to the candidate event.

[0028] In some embodiments, the generative sorting of the candidate events in the set of candidate events for the user includes:

[0029] According to at least one of the real-time scenario features of the user and the memory information for the user, the candidate events in the set of candidate events for the user are generatively sorted to obtain the recommended events for the user.

[0030] In some embodiments, the generative sorting is performed by using a generative sorting model, and an updating method of the generative sorting model includes:

[0031] According to the feedback information of the user to the recommended event, at least one of the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the recommended event is determined.

[0032] According to at least one of the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the recommended event, a reward of the recommended event is determined.

[0033] According to the reward of the recommended event, a sorting strategy of the generative sorting model is updated.

[0034] In some embodiments, the event is associated with an entry path of the intelligent agent, and the method further includes: in response to the triggering operation of the user on the event, invoking the intelligent agent and passing the solution and recommended basis information of the solution to the intelligent agent, so that the intelligent agent executes the solution, and displaying an execution result of the solution on a second interface.

[0035] Some embodiments of the present disclosure provide an information service device, comprising: one or more modules for executing an information service method.

[0036] Some embodiments of the present disclosure provide an information service device, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute an information service method based on instructions stored in the memory.

[0037] Some embodiments of the present disclosure provide a computer readable storage medium having computer instructions stored thereon, the computer instructions being executed by a processor to implement an information service method.

[0038] Some embodiments of the present disclosure provide a computer program product, comprising computer instructions, the computer instructions being executed by a processor to implement an information service method. BRIEF DESCRIPTION OF DRAWINGS

[0039] The drawings needed to be used in the following embodiments or related technical descriptions will be briefly introduced. According to the following detailed description with reference to the drawings, the present disclosure can be more clearly understood.

[0040] Obviously, the drawings in the following description are only some embodiments of the present disclosure, and other drawings can be obtained by those skilled in the art without creative labor based on these drawings.

[0041] Figure 1 A flowchart of an information service method according to some embodiments of the present disclosure is shown.

[0042] Figure 2 A flowchart of an information service method according to some embodiments of the present disclosure is shown.

[0043] Figure 3 A schematic diagram of an information service interface according to some embodiments of the present disclosure is shown.

[0044] Figure 4 A schematic diagram of deployment and update of an event reasoning model and a generative ranking model according to some embodiments of the present disclosure is shown.

[0045] Figure 5 A structural schematic diagram of an information service device according to some embodiments of the present disclosure is shown.

[0046] Figure 6A structural schematic diagram of an information service device showing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0047] It should be noted that the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure unless specifically stated otherwise.

[0048] Those skilled in the art can understand that the terms "first", "second", and the like in the embodiments of the present disclosure are only used to distinguish different steps, devices, or modules, and do not represent any specific technical meaning, nor do they represent a logical order between them.

[0049] It should also be understood that in the embodiments of the present disclosure, "multiple" can mean two or more, and "at least one" can mean one, two, or more.

[0050] It should also be understood that for any component, data, or structure mentioned in the embodiments of the present disclosure, it can be understood as one or more in general, without explicit limitation or in the context of the preceding and following text giving the opposite indication.

[0051] In addition, the term "and / or" in the present disclosure is only a description of the association relationship between the associated objects, which means that there can be three relationships, for example, A and / or B can mean that A exists alone, A and B exist together, and B exists alone. In addition, the character " / " in the present disclosure generally represents an "or" relationship between the front and rear associated objects.

[0052] It should also be understood that the description of various embodiments of the present disclosure focuses on the differences between various embodiments, and the same or similar parts can be referred to each other, and for the sake of brevity, will not be repeated one by one.

[0053] At the same time, it should be understood that, for the sake of brevity, the size of each part shown in the drawings is not drawn in accordance with the actual proportional relationship.

[0054] The following description of at least one exemplary embodiment is merely illustrative in nature and does not in any way limit the disclosure and its application or uses.

[0055] Techniques, methods, and devices known to those of ordinary skill in the relevant art can not be discussed in detail, but should be considered part of the specification where appropriate.

[0056] It should be noted that similar reference numbers and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further discussed in subsequent drawings.

[0057] Furthermore, in order not to obscure the present disclosure with unnecessary details, only the processing steps and / or device structures closely related to the solution according to the present disclosure are shown in the drawings, while other details less relevant to the present disclosure are omitted.

[0058] Figure 1 A flowchart of an information service method according to some embodiments of the present disclosure is shown. The information service method is performed by an information service device or apparatus, for example, can be performed by a front-end interaction part of the information service device or apparatus, such as a user-facing application, etc.

[0059] As shown, the information service method of the embodiments includes step 110, and can further include step 120 according to user needs. Figure 1

[0060] In step 110, in response to the first interface being triggered, information of an event recommended for the user is displayed at the first interface, the information of the event including abstract information of a solution recommended for the user and at least part of the basis information of the recommendation of the solution.

[0061] Wherein, the first interface is a user interaction interface that displays the recommended event information after being triggered, for example, can be a main interface of the information service or a main interface of an artificial intelligence service in the information service. The situation of the first interface being triggered includes, but is not limited to, user login, entering or returning to the first interface. The information service device or apparatus acquires or detects the triggering operation of the user on the first interface, and determines that the first interface is triggered.

[0062] Wherein, the event refers to a personalized, structured dynamic information unit, associated with a user-specific solution, also known as a "solution-type event". According to user-specific information (such as user behavior information, memory information for the user), a user-specific event / solution can be determined. Due to the personalized and dynamic characteristics of the user-specific information (such as user behavior information, memory information for the user), the event / solution has the characteristics of personalization and dynamics; in order to attract users, the information of the event displayed includes abstract information of a user-specific solution recommended for the user and at least part of the basis information of the recommendation of the solution, so that the event has the characteristic of being structured.

[0063] ​The solution is a solution dedicated to solving the user's needs. The summary information of the solution refers to a brief overview of the core content of the solution to solve the user's needs. The recommendation basis information of the solution refers to various types of data on which the solution is recommended. These data are initially fragmented. In this embodiment, the user is used as a traction, and in particular, the user behavior information is used as a traction. The fragmented data is associated to form an information chain with certain logical relationship as the recommendation basis information. The recommendation basis information of the solution includes, but is not limited to, the user's state, the user's potential needs, environmental information, incentives, etc., and can include one or more of them. The user's state refers to the state or attribute condition of the user at present. The user's potential needs refer to the needs that are not explicitly expressed by the user but can be inferred based on data. The environmental information refers to the external scene data of the user, including but not limited to time, space and other data. The incentive refers to the external or internal driving factor generated by the triggering event.

[0064] In order to better understand the meanings of the terms such as event, solution, and recommendation basis information, some examples are listed below. It should be understood that these examples are only used to assist in explaining the meanings of the terms, and in no way limit the meanings of the terms.

[0065] For example, the user is going to the seaside on the weekend (environment), the local forecast is that the ultraviolet rays are extremely strong (incentive), the user's historical record shows that the skin is sensitive (user state), and therefore it is inferred that the user needs a high-multiple and mild sunscreen product (solution). The information of the "event" recommended to the user displayed on the first interface is, for example, "The seaside is coming soon, and the sunscreen product is ready for you", wherein "sunscreen product" is the summary information of the user's dedicated solution, and "the seaside is coming soon" is part of the recommendation basis information of the solution, which is closely related to the user's behavior, so that the user feels that he or she is understood and enjoys the dedicated service, and the user's interactive desire is stimulated. However, according to the traditional static "thing" mode of push content, a specific sunscreen may be pushed to the user, which is relatively rigid in form and is difficult to stimulate the user's interactive desire.

[0066] For another example, the user's historical record shows that the user has just bought an air fryer (incentive), and the user is currently in a fat loss period (user state), and therefore it is inferred that the user needs a low-carb recipe air fryer usage guide (solution). The information of the "event" recommended to the user displayed on the first interface is, for example, "The air fryer has arrived, and I will teach you a few low-carb quick and easy dishes", wherein "low-carb quick and easy dishes" is the summary information of the dedicated solution, and "the air fryer has arrived" is part of the recommendation basis information of the solution, which is closely related to the user's behavior, so that the user feels that he or she is understood and enjoys the dedicated service, and the user's interactive desire is stimulated. According to the traditional static "thing" mode of push, this type of recommendation cannot be achieved.

[0067] The information service scheme of the embodiment can actively provide personalized information services in the form of "events" for the user without the user actively initiating explicit instructions, which is different from the traditional static "object" form of content. The information in the form of "events" includes abstract information of a solution tailored for the user and at least part of the recommended basis information of the solution, so that the user feels understood and enjoys exclusive services, stimulates the user's desire to interact, and improves the user's trust and use stickiness.

[0068] The type of the event may include, but is not limited to, shopping, takeout, order, article, topic, love shopping, repeat purchase, price reduction, combination purchase, usage guide, and question, and may include one or more of them.

[0069] Thus, comprehensive information services of various event types are provided to meet the full-scene needs of users in shopping, consultation, after-sales, information, etc., and to improve the functional coverage of products and user stickiness. The diversified needs of different users are adapted to, the application range of information services is widened, and the market competitiveness of products is enhanced.

[0070] The following describes various types of events. Shopping type event: an event generated around the demand for purchasing an item, for example, intelligent and personalized shopping suggestions can be generated according to user historical behavior. For example, for an office worker in a weight loss period, push "a box of low-fat instant noodles?" Takeout type event: an event generated based on takeout demand, recommend takeout-related events combined with user interests and scenarios, generate attractive slogans or questions. For example, for a person who has purchased a sandwich during a weight loss period, push "a low-fat sandwich set meal?" Order type event: an event associated with the order status of a user, showing the logistics status of all orders (from one or more applications) of the user. For example, for a user who has 3 orders (e.g., from different applications) and has viewed the order status multiple times, push "You have 3 orders recently, check the status." Article type event: an event that provides knowledge-based articles, push article titles, quick summaries, update times, etc. Special topic type event: an event that integrates content around a specific theme, push special topic titles, the latest special topic summaries, update times, etc. Love to buy type event: an exclusive event for high-frequency purchase users, push events strongly related to user purchase history. Repeat purchase type event: an event to promote user repeat purchases, for example, based on user periodic purchase behavior, remind to restock or recommend derivative categories. Discount type event: an event that pushes product discount information, monitors cart product price reductions, and pushes discount information, product selling points, etc. For example, for price-sensitive users, push "3 products you recently focused on have been reduced by up to 10%." Group purchase type event: an event that recommends group purchase of goods, for example, push "recently going skiing, ski equipment is ready for you." Usage guide type event: an event that provides usage tips or maintenance guides for purchased products, for example, for a person who has purchased an air fryer during a weight loss period, push "the air fryer has arrived, teach you a few low-carb quick recipes." Question type event: an event that answers user questions. For example, in winter, for a user who has recently consulted indoor potted plant maintenance issues, push "What should I do if the leaves of my indoor potted plant turn yellow and drop in winter?"

[0071] In the first interface, in addition to including summary information of the solution and at least part of the recommendation basis information of the solution, the information of the event can also include a reference picture of the solution, which is used to assist in displaying visual materials of the solution, to present the solution in a visual manner, reduce the understanding cost of the user for the text information, and improve the information transmission efficiency and user experience.

[0072] The reference picture source of the solution is flexible, which can come from the picture of the product related to the solution or be generated according to the solution. For example, a representative picture is selected from the picture of the product related to the solution according to a selection strategy, as the reference picture of the solution; or the solution (or the key elements of the solution) is input into a picture generation model (such as an artificial intelligence model), so that the picture generation model maps the textual elements in the solution to visual graphical elements, and generates a reference picture adapted to the solution. According to different types of solutions, the source of the reference picture is flexibly determined, the display form of the recommended event is enriched, and the attraction to the user is enhanced.

[0073] Some examples of the source of the reference picture of the solution are listed below. It should be understood that these examples are only used to assist in describing the source of the reference picture, and in no way limit the source of the reference picture.

[0074] For example, the reference picture of the solution of the event of the order type, the love purchase type, the repeat purchase type, the price reduction type, or the use guide type comes from the picture of the product related to the solution. These types of events involve information transmission and transaction conversion of actual product transactions of the user, and directly use the product picture, which can maximize the guarantee of information authenticity and persuasiveness, and more easily attract the attention of the user.

[0075] For example, the reference picture of the solution of the event of the article type, the special topic type, the combination purchase type, or the question type is generated according to the solution. These types of events involve information integration or problem solving, and there is no ready-made product picture that can be directly used, or a single product picture is not enough to represent the integrated information. In this case, the picture content can be customized according to the textual logic of the solution to meet the demand for accurate information transmission.

[0076] For example, the reference picture of the solution of the event of the shopping type or the take-out type comes from the picture of the product related to the solution or is generated according to the solution. These types of events involve complex scenes, which can involve a single type of product or multiple types of products. If a single product picture is enough to represent the overall information, the product picture can be directly used; if a single product picture is not enough to represent the overall information, the picture content can be customized according to the textual logic of the solution.

[0077] Different picture source rules are formulated for different event types to ensure that the reference picture is highly adapted to the event content and improve the rationality and effectiveness of the reference picture display.

[0078] In step 120, the event is associated with the entry path of the intelligent agent, and the intelligent agent takes over the event in response to the triggering operation of the user on the event, and displays the execution result of the solution from the intelligent agent in the second interface.

[0079] wherein the agent is a program or system with autonomous execution capability, which takes the event and outputs the execution result of the solution. The entry path is an access link associated with the event and the agent. The second interface is a user interaction interface in which the agent displays the execution result, which can be an interaction interface between the agent and the user. The trigger operation is an interactive action of the user activating the event to call the agent, including but not limited to clicking or selecting the event.

[0080] The mode of taking the event by the agent realizes the automatic execution of the solution and the feedback of the result, provides the user with exclusive information services, further stimulates the user's interactive desire, and improves the user's trust and use stickiness. However, according to the traditional way, after the user clicks the recommended content in the form of "things", it is taken by a static page, for example, it is jumped to a fixed-format / content item list page or an item detail page, which is relatively rigid in form and difficult to stimulate the user's interactive desire.

[0081] wherein the execution result of the solution is determined by the agent according to the event context. That is, when the agent is called to take the event, the event context, such as the solution, the recommended basis information of the solution, and the user information (indicating which user), is passed to the agent, so that the agent can accurately understand the user's demand, ensure that the execution result is highly matched with the user's demand, guarantee the individualization and adaptability of the execution result, avoid the execution deviation caused by the lack of information, and improve the service quality.

[0082] wherein the execution result of the solution includes a recommended item, and the recommended item includes a recommended reason, which is determined according to the solution and the recommended basis information of the user. The recommended item is the specific content output by the agent for the user to select. The recommended reason is the content based on the solution and the recommended basis, which explains the adaptability of the recommended item to the user, and is a kind of personalized recommended reason.

[0083] For example, a certain recommended item (a certain fragrance product) of a Christmas fragrance gift event: for users who like fresh and prefer cute style, the recommended reason can be "designed with cute panda as inspiration, fresh and natural with mint type, exquisite packaging, suitable as Christmas gift"; for price-sensitive and mint-preferred users, the recommended reason can be "high cost performance, unique mint type, exquisite design and packaging". Thus, different adaptability recommended reasons are designed for different users.

[0084] The recommended reason is strongly bound with the solution and the recommended basis, ensuring the rationality and pertinence of the recommendation, realizing personalized recommended reasons for the user, and making the user feel that he or she is understood and enjoys exclusive services, helping the user quickly understand the recommended value, enhancing the user's recognition of the recommended item, and improving the conversion rate.

[0085] Figure 2 A flowchart of an information service method of some embodiments of the present disclosure is shown. The information service method is performed by an information service device or apparatus, for example, can be performed by a background processing part of the information service device or apparatus, such as a server.

[0086] As shown, the information service method of this embodiment includes steps 210-230, and can also include step 240 according to user needs. Figure 2

[0087] In step 210, based on the Chain-of-Thought (CoT) capability, the candidate event set of the user is obtained by performing abductive reasoning in advance according to the user's behavior information and the memory information for the user.

[0088] This step can perform offline deep event reasoning (slow thinking), and can use non-real-time powerful computing power to solve the depth problem of event reasoning.

[0089] Among them, the Chain-of-Thought (CoT) capability is the ability to derive user needs and events through multi-step reasoning. Abductive reasoning is a reasoning method that reverses the reason based on the result to predict user needs. The user's behavior information is the operation data of the user on various platforms / applications. The memory information for the user is the user's characteristics mined at different times.

[0090] In addition to user-specific information such as user's behavior information and memory information for the user, world knowledge can also be combined to perform abductive reasoning to obtain the candidate event set of the user. World knowledge refers to the sum of facts, laws, connections and cognition about geography, history, culture, politics, economy, society, technology and other multi-disciplinary fields in the global range. Thus, the candidate events reasoned are not only in line with the user's personalized needs, but also in line with objective laws and general scene logic, greatly improving the accuracy, rationality and practicality of the events.

[0091] Among them, the memory information for the user can include memory information at different times, for example, including: the user's preferences in the first period, the user's state in the second period, and the user's intention in the third period, wherein the first period is greater than the second period, and the second period is greater than the third period. The memory information of the first period is also called long-term memory, which captures the user's fixed characteristics; the memory information of the second period is also called medium-term memory, which captures the user's stage state; and the memory information of the third period is also called short-term memory, which captures the user's short-term interest fluctuations. For example, the first period is in units of months, the second period is in units of weeks, and the third period is in units of days, but is not limited to the examples shown, and the length of each period can be flexibly set according to business needs. ​

[0092] By using long / medium / short-term layered memory, the characteristics of the user at different periods are comprehensively captured, the long-term preferences / habits, the medium-term state and the short-term intention of the user are considered, and multi-dimensional accurate profiling is realized. Based on the inference event of layered memory, the recommendation is ensured to conform to the long-term preferences of the user, and the current state and the immediate intention are adapted, and the comprehensive adaptation ability of the event recommendation is improved.

[0093] The memory information for the user includes a memory slice for the user and context logic formed by the memory slice. Each memory slice refers to a feature of the user. For example, the user features depicted in the long-term memory include “minimalist” and “fitness enthusiast”, which are two memory slices; the user features depicted in the medium-term memory include “under renovation”, which is also a memory slice. The context logic formed by the memory slice refers to the logical relationship of inferring the user features based on the user behavior. For example, the user often exercises and purchases a large number of exercise equipment, and thus infers that the user is a “fitness enthusiast”. The basis information and inference result of this inference can be used as the context logic of the “fitness enthusiast” memory slice.

[0094] The context logic of the memory slice is retained to provide complete user behavior link basis for event inference, and to avoid event inference deviation caused by information fragmentation. Based on the context logic, the correlation of the user demand can be mined, the event more consistent with the user behavior habit is generated, and the coherence and rationality of the recommendation are improved.

[0095] In some embodiments, a backward reasoning can be performed by using an event inference model, which may, for example, include but is not limited to a large language model (LLM). The large language model is a deep learning model trained using a large amount of data. Various data sources, user behavior information, memory information for the user, and world knowledge, etc. are input into the event inference model, the event inference model reverses the “cause” (event) from the “effect” (behavior fragment), and outputs a candidate event set of the user.

[0096] An exemplary reasoning process of the event inference model includes: (1) phenomenon induction: it is observed that the user looked at a tent and searched for a barbecue grill; (2) hypothesis: the user may go camping; (3) conflict detection and verification (counterfactual reasoning): combined with the weather forecast, it is found that it will rain this weekend, and the inference is corrected to: the user may be looking for an alternative solution suitable for indoor or postponing the plan; (4) event instantiation: an event entity of “rainy day alternative camping plan” is generated.

[0097] In step 220, in response to the first interface being triggered, the candidate events in the candidate event set of the user are generated and ranked, and an event recommended for the user is obtained.

[0098] The step can be performed online generation ranking (fast decision), and under strict online delay constraints, from a high-quality candidate event set generated by near-line generation, the event that can most accurately solve the user's problem at the current time is matched through the generation ranking strategy, and online fast response is realized.

[0099] Through the two-stage recommendation framework of slow thinking event generation and fast decision event selection, the contradiction between event reasoning depth and real-time response speed in the event recommendation process is effectively alleviated.

[0100] Among them, the generation ranking is a way of dynamically generating ranking results based on real-time and historical data. For example, according to at least one of the real-time scenario features of the user and the memory information for the user, the candidate events in the candidate event set of the user are ranked to obtain the events recommended for the user. Among them, the memory information can be short-term memory, and can also include medium-term memory and long-term memory.

[0101] Among them, the real-time scenario feature is the current real-time micro-situation vector collected when the user opens the recommendation application or triggers interaction, for example, including but not limited to environmental features, conversation features, and immediate behavior flow. Environmental features: such as device type, current time slice (such as "morning rush hour on weekdays"), geographic location (Point of Information (POI), such as "in the hospital" or "in the mall"), network environment. Conversation features: entry source (home page / message center / activity page), last hop page identifier, current stay duration. Immediate behavior flow: user's recent time (such as within 5 minutes) click sequence or search word (as the strongest signal of immediate intent).

[0102] The real-time scenario feature is integrated during ranking, so that the recommended event can match the current real-time scenario of the user, improving the immediate adaptability of the event; the memory information is integrated during ranking, covering the user's long / medium / short-term preferences, state and intent, which can avoid the recommendation deviation caused by relying only on real-time behavior (such as accidental clicks). The real-time scenario feature and the memory information are integrated for ranking to find the event that can most solve the user's needs at the current time.

[0103] In some embodiments, generation ranking can be performed using a generation ranking model, which for example includes but is not limited to a pre-trained or distilled lightweight large language model to ensure online reasoning speed. The real-time scenario features of the user and the memory information for the user and the key metadata of the candidate event set are organized into a model input sequence, which is input into the generation ranking model. The generation ranking model directly outputs the sorted event ID sequence in a sequence generation manner, or directly outputs a number of event IDs with high confidence ranking.

[0104] In addition, the event generated by the generative ranking model can be subjected to business rule verification, and finally the valid event can be output and displayed to the user. The business rule verification includes, for example, recommendation frequency control, life cycle check, inventory status check, etc. Through the recommendation frequency control, events or recommended items with too high recommendation frequency are filtered out. Through the life cycle check, expired recommended items (such as goods, information) are filtered out. Through the inventory status check, recommended items (such as goods) that are not currently in stock are filtered out.

[0105] In step 230, the information of the event is displayed on the first interface, the information of the event including the summary information of the solution recommended for the user and at least part of the recommendation basis information of the solution.

[0106] Therefore, without the user actively initiating an explicit instruction, the user can be actively provided with personalized information services in the form of an "event", which is different from traditional static content in the form of a "thing". The information in the form of an "event" includes the summary information of a "tailor-made" solution for the user and at least part of the recommendation basis information of the solution, so that the user feels understood and enjoys exclusive services, stimulates the user's interactive desire, and improves the user's trust and use stickiness.

[0107] In step 240, the event is associated with an entry path of an intelligent agent, and in response to the triggering operation of the user on the event, the intelligent agent is called to execute the solution, and the execution result of the solution is displayed on the second interface.

[0108] The mode of adopting the intelligent agent to undertake the event realizes the automatic execution and result feedback of the solution, provides exclusive information services for the user, further stimulates the user's interactive desire, and improves the user's trust and use stickiness.

[0109] When the intelligent agent is called, the context of the event is passed to the intelligent agent, and the context of the event includes, for example, the solution, the recommendation basis information of the solution, the user information (indicating which user), etc. The intelligent agent executes the solution according to the context of the event to obtain the execution result of the solution, so that the intelligent agent can accurately understand the user's demand, ensure that the execution result is highly matched with the user's demand, guarantee the personalization and adaptability of the execution result, avoid the execution deviation caused by the lack of information, and improve the service quality.

[0110] Figure 3 A schematic diagram of an information service interface showing some embodiments of the present disclosure. As shown in FIG. 1, the information service interface includes a first interface and a second interface. Figure 3As shown, four events of multiple types are recommended and displayed on the first interface for a certain user, and the information of each event shows the summary of the user-specific solution and the basis for the recommendation. These events can be of the same type or of different types. Assuming that the user is interested in event 2, after triggering event 2, the intelligent agent takes over event 2 and enters the second interface displayed by the intelligent agent. The second interface displays the execution result of solution 2, including three recommended items, each of which has its own personalized recommendation reason. In addition, the intelligent interaction area is also set on the first interface and the second interface, and the user can directly interact with the intelligent agent.

[0111] Figure 4 Deployment and update schematic diagram of the event reasoning model and the generative ranking model of some embodiments of the present disclosure.

[0112] Various behavior information of the user is obtained from the original log, and memory information for the user is mined based on the various behavior information of the user, such as long-term / medium-term / short-term hierarchical memory. Various data sources, the behavior information of the user, the memory information for the user, and world knowledge, etc. are input into the event reasoning model, and the event reasoning model performs abductive reasoning to output a candidate event set of the user.

[0113] When the user logs in, enters, or returns to the first interface, in response to the first interface being triggered, the real-time context features of the user, the memory information for the user, and the key metadata of the candidate event set, etc. are organized into a model input sequence, which is input into the generative ranking model. The generative ranking model sorts the candidate events in the candidate event set of the user in a sequence generation manner to obtain a personalized / specific event recommended for the user, outputs a sequence of sorted event IDs, or directly outputs several event IDs with high confidence ranking, and displays the information of the recommended event on the first interface.

[0114] The user browses the recommended event on the first interface, and may only browse, click, or close the event, etc. These operations as user feedback can reflect the degree of interest of the user in the event. The feedback of the user to the displayed event is recorded, including explicit feedback, implicit feedback, etc., to form training data. The explicit feedback includes, for example: clicking, closing, negative feedback ("not interested"). The implicit feedback includes, for example: the length of stay in the event, whether to enter the downstream intelligent agent session, and the completion rate of multiple event concatenation.

[0115] According to the feedback information of the user to the recommended event, the event reasoning model and the generative ranking model are updated to optimize the event reasoning strategy and the event ranking strategy.

[0116] The event reasoning model is updated by using an RL (Reinforcement Learning) method, including but not limited to updating the event reasoning model strategy by using a GRPO (Group Relative Policy Optimization), that is, using a multi-path sampling (group sampling) manner for the same user context, and using the relative advantages of multiple candidate events generated in the group to optimize the event reasoning strategy.

[0117] The group of candidate events generated for the same user context (same behavior + memory information) is evaluated by relative comparison in the group instead of isolated absolute reward evaluation, and the interference of reward difference in different user contexts on model updating is eliminated. The strategy is updated based on the relative advantages in the group, so that the model clearly knows "which events are more competitive in the same user context", and forces the model to iterate in the direction of "generating high relative advantage events".

[0118] Based on the GRPO, the updating method of the event reasoning model includes the following steps.

[0119] (1) For a group of candidate events generated based on the same behavior information and memory information of the user (i.e., the same user context) in the candidate event set, the reward of each candidate event in the group is calculated.

[0120] The calculation of the reward of each candidate event in the group includes: calculating the reward of each candidate event in the group according to at least one of the logical self-consistency, surprise degree, format, interaction rate, task completion rate, retention value, and disturbance penalty value of each candidate event in the group, wherein the interaction rate, task completion rate, retention value, and disturbance penalty value of the candidate event are determined according to the feedback information of the user to the candidate event.

[0121] When calculating the reward, the relevant items of the reward are weighted and summed to obtain the reward of the event. The formula can be expressed as, for example: R = w1 x R1 + w2 x R2 + w3 x R3 + w4 x CTR + w5 x CVR + w6 x LTV - w7 x Dis, wherein R represents the reward of the event, R1 represents the logical self-consistency of the event, R2 represents the surprise degree of the event, R3 represents the format evaluation of the event, CTR represents the interaction rate of the event, CVR represents the task completion rate of the event, LTV represents the retention value of the event, Dis represents the disturbance penalty value of the event, and w1, w2, w3, w4, w5, w6, and w7 represent the weighting coefficients of the respective relevant items.

[0122] Logical self-consistency is used to represent whether the matching logic of the event and the user context (behavior information + memory information) is rigorous, reasonable, and contradictory. It can be manually evaluated or evaluated by a model. When the model is evaluated, the model can be used to match the similarity between the description text of the event and the user context features. The greater the similarity, the better the logical self-consistency. The logical self-consistency is included in the reward dimension, which forces the logical matching degree of the candidate event and the user context, and avoids generating illusion events that violate the user's demand and common sense.

[0123] Surprise degree represents the degree to which the event exceeds the user's historical interaction expectations, reflecting the novelty and potential value of the event, and avoiding recommending mediocre content. It can be manually evaluated or evaluated by a model. When the model is evaluated, the event set of the user's historical interaction can be counted, and the similarity between the current event and the events in the set is calculated. The lower the similarity, the higher the surprise degree. The surprise degree is included in the reward dimension, which forces the model to dig into the user's potential demand and avoids repeatedly generating the same events as the user's historical interaction (mediocrity).

[0124] Combining logical self-consistency and surprise degree, the illusion and mediocrity problems in the event reasoning process of the model are solved.

[0125] Format represents whether the information display format of the event conforms to the preset specification (such as title structure, abstract length, picture collocation, etc.), reflecting the readability of the event. It can be manually evaluated or evaluated by a model. When the model is evaluated, the format compliance can be scored based on the preset format compliance verification rules. The higher the score, the more standardized the format.

[0126] Interaction rate represents whether the user clicks or views the event, reflecting the attractiveness of the event to the user. The higher the interaction rate, the stronger the user's interaction willingness.

[0127] Task completion rate represents the proportion of users completing the target task guided by the event after triggering the event, reflecting the accuracy of the event in solving the user's problem.

[0128] Retention value represents the contribution of the event to improving the user's subsequent platform / application retention, reflecting the long-term value of the event. For example, the retention value is calculated based on the user's revisit to the platform / application within a preset number of days after interacting with the event.

[0129] Disturbance penalty value represents the degree of disturbance caused by the event to the user, which is counted into the reward in reverse. For example, the higher the disturbance penalty value, the more the event disturbs the user, such as the user quickly swiping or clicking to close or clicking the event that is not interested.

[0130] The multi-dimensional reward system combines "logical self-consistency (reasonableness), surprise degree (novelty), format (standardization), interaction rate / task completion rate (practicality), retention value (long-term value), and disturbance penalty value (friendliness)". It not only ensures the basic quality of the event, but also takes into account user experience and business value, avoiding the model from falling into the trap of "sensationalism" and "low-quality content" due to single-dimensional orientation (such as only pursuing click-through rate).

[0131] Moreover, the four dimensions of interaction rate, task completion rate, retention value, and disturbance penalty value are driven by user feedback, making the reward calculation strongly bound to the user's real experience. User behaviors such as clicking, task completion, revisiting, and negative feedback are converted into signals for model updates in real time, promoting continuous iteration and optimization of the model, and achieving a virtuous cycle of "user demand-event generation-feedback optimization".

[0132] (2) According to the reward of each candidate event in the group, calculate the relative advantage of each candidate event in the group.

[0133] For example, based on the reward of each candidate event in the group, calculate the reward average of each candidate event in the group, and determine the relative advantage of a single candidate event in the group according to the difference (or normalized difference) between the reward of a single candidate event and the reward average.

[0134] (3) Update the reasoning strategy of the event reasoning model according to the relative advantages of all candidate events in the group.

[0135] According to the relative advantages of each candidate event in the group and the generation probability of the candidate event, construct a target function to maximize the generation probability of high-advantage candidate events in the group, and update the event reasoning model parameters based on the target function using gradient descent to optimize the reasoning strategy of the event reasoning model. The event reasoning model can be updated periodically (such as weekly / monthly).

[0136] Thus, the relative advantage output by the group is used to update the strategy model, enabling the model to generate events with more rigorous logic and more insightful insights.

[0137] For the generative ranking model, a reward function is specially designed for "decision quality", not only focusing on clicks, but also focusing on the virtuous cycle of interaction. The update method of the generative ranking model includes the following steps.

[0138] (1) According to the feedback information of the recommended event by the user, determine at least one of the interaction rate, task completion rate, retention value, and disturbance penalty value of the recommended event. The meanings and calculation methods of each item are described above and will not be repeated here.

[0139] (2) Determine the reward of the recommended event according to at least one of the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the recommended event.

[0140] In the calculation of the reward, the relevant items are weighted and summed to obtain the reward of the event. The formula can be expressed as, for example: R = w1 x CTR + w2 x CVR + w3 x LTV - w4 x Dis, where R represents the reward of the event, CTR represents the interaction rate of the event, CVR represents the task completion rate of the event, LTV represents the retention value of the event, Dis represents the disturbance penalty value of the event, and w1, w2, w3, and w4 represent the weighting coefficients of the respective relevant items.

[0141] (3) Update the ranking strategy of the generative ranking model according to the reward of the recommended event.

[0142] The reinforcement learning method is used to update the generative ranking model, including but not limited to using the GRPO algorithm or PPO (Proximal Policy Optimization) and the like. Based on the reward of the event, the generative ranking model is updated periodically (such as hourly / daily).

[0143] Based on the reward of the event and the ranking probability (i.e., the recommendation probability) of the event, a target function is constructed to maximize the ranking probability of high-reward events. Based on the target function, the generative ranking model parameters are updated using gradient descent to optimize the ranking strategy of the generative ranking model.

[0144] Thus, the generative ranking model can learn to preferentially display high-value events that can stimulate immediate clicks and promote long-term interactive interactions, rather than just displaying highly inducible "sensationalist" content when faced with complex micro-situations.

[0145] The two-stage event reasoning system based on the large language model cognitive architecture of the embodiments of the present disclosure upgrades the traditional "content recommendation" to "life event reasoning". The system not only distributes information, but also infers the specific life event of the user based on user behavior, memory, and world knowledge, and initiates high-value density interactions as a carrier, thereby significantly improving the user opening rate and user trust of interactive artificial intelligence products.

[0146] The information service method of the embodiments of the present disclosure can actively recommend users in the aspects of commodities, take-out, information, topics, travel, cleaning, logistics, and vehicle use based on user data, and the dialogue interface of the agent can be used after clicking. The scenarios that can be used include, but are not limited to, shopping decision, regular purchase, discount exploration, post-purchase knowledge, and logistics tracking. The data sources for recommendation include, but are not limited to, various orders, additions, searches, browsing, collections, attentions, and dialogue records of users. The home page content is refreshed every day when the user enters the application or a new dialogue is created, and the user is not repeatedly recommended to give the user more freshness.

[0147] Figure 5 A structural schematic diagram of an information service device of some embodiments of the present disclosure is shown. The information service device includes one or more modules that perform an information service method.

[0148] As shown in Figure 5 The information service device 500 of the embodiments includes the module 510 and can further include the modules 520-540.

[0149] The display module 510 is configured to display, in response to a first interface being triggered, information of an event recommended for a user in the first interface, the information of the event including abstract information of a dedicated solution recommended for the user and at least part of recommendation basis information of the solution.

[0150] The recommendation basis information of the solution includes one or more of the state of the user, potential needs of the user, environmental information, and incentives.

[0151] The event is associated with an entry path of an agent. The display module 510 is configured to display, in response to a triggering operation of the event by the user, an execution result of the solution from the agent in a second interface.

[0152] The execution result of the solution includes a recommended item, and the recommended item includes a recommendation reason determined according to the solution and the recommendation basis information of the user.

[0153] The execution result of the solution is obtained by the agent executing the solution according to the recommendation basis information of the solution.

[0154] The information of the event includes a reference picture of the solution, and the reference picture is from a picture of a product related to the solution or generated according to the solution.

[0155] The type of the event includes one or more of a shopping type, a take-out type, an order type, an article type, a special topic type, a love purchase type, a repeat purchase type, a price reduction type, a combination purchase type, a use guide type, and a question type.

[0156] The reference picture of the solution of the event of the order type, the love purchase type, the repeat purchase type, the price reduction type, or the use guide type: a picture from a product related to the solution; or the reference picture of the solution of the event of the article type, the special topic type, the combination purchase type, the question type: generated according to the solution; or the reference picture of the solution of the event of the shopping type, the take-out type: a picture from a product related to the solution or generated according to the solution.

[0157] The reasoning module 520 is configured to perform a prior abductive reasoning based on a thinking chain capability according to the behavior information of the user and the memory information for the user, to obtain a candidate event set of the user;

[0158] The sorting module 530 is configured to, in response to the first interface being triggered, perform generative sorting on the candidate events in the candidate event set of the user, to obtain an event recommended for the user; and display information of the event on the first interface.

[0159] The memory information for the user includes a memory slice for the user and a context logic formed by the memory slice.

[0160] The memory information for the user includes a preference of the user in a first period, a state of the user in a second period, and an intention of the user in a third period, where the first period is greater than the second period, and the second period is greater than the third period.

[0161] The sorting module 530 is configured to perform generative sorting on the candidate events in the candidate event set of the user according to at least one of the real-time scene feature of the user and the memory information for the user, to obtain an event recommended for the user.

[0162] The model updating module 540 is configured to perform abductive reasoning by using an event reasoning model, and an updating method of the event reasoning model includes:

[0163] For a group of candidate events in the candidate event set, which are generated based on the same behavior information and memory information of the user, a reward of each candidate event in the group is calculated;

[0164] According to the reward of each candidate event in the group, a relative advantage of each candidate event in the group is calculated;

[0165] According to the relative advantages of all candidate events in the group, an inference strategy of the event reasoning model is updated.

[0166] The calculating the reward of each candidate event in the group comprises: calculating the reward of each candidate event in the group according to at least one of logical self-consistency, surprise degree, format, interaction rate, task completion rate, retention value, and disturbance penalty value of each candidate event in the group, wherein the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the candidate event are determined according to the feedback information of the user to the candidate event.

[0167] The model updating module 540 is configured to perform generative ranking by using a generative ranking model, and an updating method of the generative ranking model comprises:

[0168] determining at least one of the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the recommended event according to the feedback information of the user to the recommended event;

[0169] determining the reward of the recommended event according to at least one of the interaction rate, the task completion rate, the retention value, and the disturbance penalty value of the recommended event;

[0170] updating a ranking strategy of the generative ranking model according to the reward of the recommended event.

[0171] The display module 510 is configured to, in response to the triggering operation of the user to the event, invoke the agent and pass the solution and the recommended basis information of the solution, so that the agent executes the solution, and display an execution result of the solution on a second interface.

[0172] Figure 6 A structural schematic diagram of an information service device of some embodiments of the present disclosure is shown. As shown in the figure, the information service device 600 of the embodiment comprises a memory 610 and a processor 620 coupled to the memory 610, and the processor 620 is configured to execute the information service method in any of the foregoing embodiments based on instructions stored in the memory 610. For details, please refer to the foregoing, which will not be repeated here. Figure 6

[0173] The information service device 600 can further comprise an input / output interface 630, a network interface 640, a storage interface 650, etc. These interfaces 630, 640, 650 and the memory 610 and the processor 620 can be connected through a bus 660, for example.

[0174] The memory 610 can comprise a system memory, a fixed non-volatile storage medium, etc. The system memory stores, for example, an operating system, an application program, a boot loader (Boot Loader), and other programs, etc.

[0175] ​The processor 620 can be implemented in a manner of a general processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, a discrete gate or transistor, etc.

[0176] The input / output interface 630 provides a connection interface for input / output devices such as a display, a mouse, a keyboard, a touch screen, etc. The network interface 640 provides a connection interface for various networking devices. The storage interface 650 provides a connection interface for external storage devices such as an SD card, a U disk, etc. The bus 660 can use any bus structure in a plurality of bus structures. For example, the bus structure includes but is not limited to an industry standard architecture (ISA) bus, a micro channel architecture (MCA) bus, a peripheral component interconnect (PCI) bus.

[0177] In the technical solutions of the present disclosure, the collection, collection, updating, analysis, processing, use, transmission, storage, etc. of user personal information involved in the technical solutions comply with relevant laws and regulations, are used for legal purposes, do not violate public order and good customs, and need to inform the user and obtain the user's consent or authorization, when applicable, the user's personal information is de-identified and / or anonymized and / or encrypted. Technical processing. Necessary measures are taken on user personal information to prevent illegal access to user personal information data, and to maintain user personal information security, network security and national security.

[0178] Those skilled in the art should understand that the embodiments of the present disclosure can be provided as a method, a system, or a computer program product. Therefore, the present disclosure can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present disclosure can take the form of a computer program product implemented on one or more (non-transitory) computer-readable storage media containing computer program code (including but not limited to disk storage, CD-ROM, optical storage, cloud storage, etc.). The computer program product should be understood as a software product that mainly realizes its solutions through a computer program.

[0179] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram.

[0180] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram.

[0181] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram. Figure 1 one or more flows and / or blocks in the flowchart and / or block diagram.

Claims

1. An information service method, comprising: In response to the first interface being triggered, information about an event recommended to the user is displayed on the first interface. The event information includes a summary of the personalized solution recommended to the user and at least part of the basis for the recommendation of the solution.

2. The information service method according to claim 1, wherein, The information used to recommend the solution includes one or more of the following: the user's status, the user's potential needs, environmental information, and incentives.

3. The information service method according to claim 1, wherein, The event is associated with the agent's entry path and also includes: In response to the user's triggering operation on the event, the execution result of the solution from the intelligent agent is displayed on the second interface.

4. The information service method according to claim 3, wherein, The execution result of the solution includes recommended items, and the recommended items include a recommendation reason, which is determined based on the user's solution and the recommendation basis information.

5. The information service method according to claim 3, wherein, The execution result of the solution is obtained by the agent executing the solution based on the recommended information of the solution.

6. The information service method according to claim 1, wherein, The information about the event includes reference images of the solution, which are derived from images of products related to the solution or generated based on the solution.

7. The information service method according to claim 1, wherein, The types of events include one or more of the following: shopping, food delivery, order, article, topic, favorite shopping, repeat purchase, price reduction, bundled purchase, user guide, and question.

8. The information service method according to claim 7, wherein: Reference images for solutions to events related to order type, favorite purchase type, repeat purchase type, price reduction type, or user guide type: images from products related to the stated solution; or, Reference images for solutions to events of article type, topic type, bundled purchase type, and question type: generated based on the stated solutions; or, Reference images for solutions to shopping and food delivery type events: images from products related to the solution or generated based on the solution.

9. The information service method according to any one of claims 1-8, further comprising: Based on the thinking chain capability, abductive reasoning is performed in advance according to the user's behavioral information and memory information about the user to obtain the user's candidate event set; In response to the first interface being triggered, the candidate events in the user's candidate event set are generatively sorted to obtain the events recommended for the user; The information about the event is displayed on the first interface.

10. The information service method according to claim 9, wherein, The memory information for the user includes: memory slices for the user and the context logic formed by the memory slices.

11. The information service method according to claim 9, wherein, The memory information for the user includes: the user's preferences in a first period, the user's state in a second period, and the user's intent in a third period, wherein the first period is longer than the second period, and the second period is longer than the third period.

12. The information service method according to claim 9, wherein, Abductive reasoning is performed using an event-based reasoning model, and the method for updating the event-based reasoning model includes: For a set of candidate events generated based on the same behavioral and memory information of the user from the candidate event set, calculate the reward for each candidate event in the set; Calculate the relative advantage of each candidate event within the group based on the reward of each candidate event within the group; The reasoning strategy of the event reasoning model is updated based on the relative advantage of all candidate events within the group.

13. The information service method according to claim 12, wherein, The reward for each candidate event within the calculation group includes: The reward for each candidate event in the group is calculated based on at least one of the following: logical consistency, surprise factor, format, interaction rate, task completion rate, retention value, and disturbance penalty value. The interaction rate, task completion rate, retention value, and disturbance penalty value of the candidate event are determined based on the user's feedback information on the candidate event.

14. The information service method according to claim 9, wherein, Generatively sorting is performed on the candidate events in the user's candidate event set to obtain the events recommended for the user, including: Based on at least one of the user's real-time contextual characteristics and memory information about the user, the candidate events in the user's candidate event set are generatively sorted to obtain the events recommended for the user.

15. The information service method according to claim 9, wherein, Generative sorting is performed using a generative sorting model, and the update method of the generative sorting model includes: Based on the user's feedback on the recommended event, determine at least one of the following: interaction rate, task completion rate, retention value, and disturbance penalty value of the recommended event; The reward for the recommended event is determined based on at least one of the following: interaction rate, task completion rate, retention value, and disturbance penalty value. The ranking strategy of the generative ranking model is updated based on the reward for the recommended event.

16. The information service method according to claim 9, wherein, The event is associated with the agent's entry path and also includes: In response to the user's triggering operation on the event, the agent is invoked, and the solution and the recommendation basis information of the solution are passed to the agent so that the agent can execute the solution and display the execution result of the solution on the second interface.

17. An information service device, comprising: One or more modules that perform the information service method according to any one of claims 1-16.

18. An information service device, comprising: Memory; And a processor coupled to the memory, the processor being configured to execute the information service method of any one of claims 1-16 based on instructions stored in the memory.

19. A computer-readable storage medium having stored thereon computer instructions that, when executed by a processor, implement the information service method of any one of claims 1-16.

20. A computer program product comprising computer instructions that, when executed by a processor, implement the information service method according to any one of claims 1-16.

Images