Transaction recommendation method and apparatus, device, and storage medium

Abstract

Embodiments of the present specification provide a transaction recommendation method, device and equipment and a storage medium. The method comprises: obtaining a transaction consultation message of a target object, determining an action corresponding to the transaction consultation message based on the transaction consultation message, a plurality of historical transaction consultation messages and an object feature of the target object, wherein the transaction consultation message and the plurality of historical transaction consultation messages can reflect the latest demand of the target object, and the object feature can reflect the characteristics of the target object. The determined action also meets the personalized demand of the target object. In the case of recommending a transaction as the action, the target transaction is determined from a plurality of candidate transactions based on the transaction consultation message, and the target transaction is recommended to the target object.

Description

Technical Field

[0001] This specification relates to the field of computer technology, and in particular to a transaction recommendation method, apparatus, device, and storage medium. Background Technology

[0002] With the development of computer technology, many transaction platforms use computer technology to recommend transactions to users. For example, many insurance platforms use computer technology to recommend insurance products to users.

[0003] In related technologies, transaction platforms recommend new transactions to users based on their past transaction history. For example, an insurance platform recommends new insurance products based on previously purchased insurance products. However, since users' needs for transactions are not static, recommending new transactions based on past transactions may not match the user's needs, resulting in poor transaction recommendation effectiveness. Summary of the Invention

[0004] This specification provides a transaction recommendation method, apparatus, device, and storage medium, which can improve the transaction recommendation effect. The technical solution is as follows:

[0005] On the one hand, a transaction recommendation method is provided, the method comprising:

[0006] Retrieve transaction consultation messages from the target object;

[0007] Based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object, the action corresponding to the transaction consultation message is determined, and the action includes replying to the transaction consultation message and recommending a transaction.

[0008] In the case where the action is a recommended transaction, the target transaction is determined from multiple candidate transactions based on the transaction consultation message, wherein the multiple candidate transactions are determined based on the multiple historical transaction consultation messages;

[0009] The target transaction is recommended to the target object.

[0010] On the one hand, a transaction recommendation device is provided, the device comprising:

[0011] The message acquisition module is used to acquire transaction consultation messages from the target object;

[0012] The action determination module is used to determine the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object. The action includes replying to the transaction consultation message and recommending a transaction.

[0013] The transaction determination module is used to determine the target transaction from multiple candidate transactions based on the transaction consultation message when the action is a recommended transaction. The multiple candidate transactions are determined based on the multiple historical transaction consultation messages.

[0014] The recommendation module is used to recommend the target transaction to the target object.

[0015] In one possible implementation, the action determination module is used to determine the state of the target object based on multiple historical transaction consultation messages of the target object and the object characteristics of the target object; and to determine the action corresponding to the transaction consultation message based on the state of the target object and the transaction consultation message.

[0016] In one possible implementation, the action determination module is used to input multiple historical transaction consultation messages of the target object and the object features of the target object into a state determination model; through the state determination model, feature extraction is performed on the multiple historical transaction consultation messages to obtain the context features of the multiple historical transaction consultation messages; through the state determination model, the context features of the multiple historical transaction consultation messages and the object features of the target object are fused to output the state of the target object.

[0017] In one possible implementation, the action determination module is used to input the state of the target object and the transaction consultation message into the action determination model; through the action determination model, the state of the target object and the transaction consultation message are mapped, and the action corresponding to the transaction consultation message is output.

[0018] In one possible implementation, the transaction determination module is configured to extract interests from the transaction consultation message to obtain interest points in the transaction consultation message; and to determine the target transaction from the plurality of candidate transactions based on the interest points in the transaction consultation message.

[0019] In one possible implementation, the transaction determination module is used to perform tag mapping on the points of interest in the transaction consultation message to obtain transaction tags corresponding to the points of interest in the transaction consultation message; and to use the transaction tags to filter among the multiple candidate transactions to obtain the target transaction, wherein the target transaction has the transaction tags.

[0020] In one possible implementation, the recommendation module is configured to perform any of the following:

[0021] Send the interaction card for the target transaction to the terminal used by the target object;

[0022] The media resources of the target transaction are sent to the terminal used by the target object, and the media resources are used to introduce the target transaction;

[0023] Send a virtual object control instruction to the terminal used by the target object, the virtual object control instruction being used to instruct the terminal to control the virtual object to introduce the target transaction.

[0024] In one possible implementation, the device further includes:

[0025] The object feature acquisition module is used to acquire object information of the target object and interaction behavior information on the session interface; and to determine the object features of the target object based on the object information of the target object and the interaction behavior information.

[0026] In one possible implementation, the device further includes:

[0027] The recall module is used to extract interests from the multiple historical transaction consultation messages to obtain multiple points of interest in the multiple historical transaction consultation messages; and to recall the multiple candidate transactions based on the multiple points of interest in the multiple historical transaction consultation messages.

[0028] In one possible implementation, the recall module is used to perform tag mapping on the plurality of points of interest to obtain a plurality of transaction tags corresponding to the plurality of points of interest; and to match the plurality of transaction tags with a plurality of preset transactions to obtain the plurality of candidate transactions.

[0029] In one possible implementation, the device further includes:

[0030] The structuring module is used to structure the introductory text of multiple preset transactions to obtain the multiple preset transaction tags.

[0031] In one possible implementation, the device further includes:

[0032] The intent recognition module is used to perform intent recognition on the multiple historical transaction consultation messages to obtain multiple intents corresponding to the multiple historical transaction consultation messages; if any of the multiple intents is the target intent, the module performs interest extraction on the multiple historical transaction consultation messages to obtain multiple points of interest in the multiple historical transaction consultation messages.

[0033] In one possible implementation, the device further includes:

[0034] The message determination module is used to determine the transaction tag corresponding to the transaction inquiry message when the action is to reply to the transaction inquiry message; and to determine the reply message to the transaction inquiry message based on the transaction tag.

[0035] On one hand, a computer device is provided, the computer device including one or more processors and one or more memories, the one or more memories storing at least one computer program, the computer program being loaded and executed by the one or more processors to implement the transaction recommendation method.

[0036] On one hand, a computer-readable storage medium is provided, wherein at least one computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the transaction recommendation method.

[0037] On one hand, a computer program product or computer program is provided, which includes program code stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the aforementioned transaction recommendation method.

[0038] The technical solution provided in the embodiments of this specification obtains transaction consultation messages for a target object. Based on these messages, multiple historical transaction consultation messages, and the object characteristics of the target object, the corresponding action is determined. The transaction consultation message and the historical transaction consultation messages reflect the latest needs of the target object, and the object characteristics reflect the object's features. Therefore, the determined action aligns with the target object's personalized needs. When the action is a recommended transaction, the transaction consultation message is used to filter and determine the target transaction from multiple candidate transactions. This target transaction is then recommended to the target object, as it better meets the target object's needs. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a schematic diagram of the implementation environment of a transaction recommendation method provided in the embodiments of this specification;

[0041] Figure 2 This is a flowchart of a transaction recommendation method provided in the embodiments of this specification;

[0042] Figure 3 This is a flowchart of another transaction recommendation method provided in the embodiments of this specification;

[0043] Figure 4 This is a schematic diagram of the interface provided in the embodiments of this specification;

[0044] Figure 5 This is a schematic diagram of a system provided in an embodiment of this specification;

[0045] Figure 6 This is a schematic diagram of the structure of a transaction recommendation device provided in the embodiments of this specification;

[0046] Figure 7 This is a schematic diagram of the structure of a terminal provided in an embodiment of this specification;

[0047] Figure 8 This is a schematic diagram of the structure of a server provided in the embodiments of this specification. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments in this specification clearer, the implementation methods of this specification will be further described in detail below with reference to the accompanying drawings.

[0049] In this manual, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0050] In order to clearly explain the technical solutions provided in the embodiments of this specification, the terms involved in the embodiments of this specification will be introduced below.

[0051] Artificial intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence.

[0052] Machine Learning (ML) is a multidisciplinary field that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, and many other disciplines. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge sub-models to continuously improve their performance.

[0053] Semantic features: Features used to represent the semantics expressed by text. Different texts can correspond to the same semantic features; for example, the text "How's the weather today?" and the text "How's the weather today?" can correspond to the same semantic feature. Computer devices can map characters in text into character vectors, and combine and operate on these character vectors according to the relationships between characters to obtain the semantic features of the text. For example, computer devices can use a bidirectional encoder representation from transformers (BERT).

[0054] Normalization: Mapping sequences of values ​​with different ranges to the interval (0, 1) to facilitate data processing. In some cases, normalized values ​​can be directly expressed as probabilities.

[0055] Dropout is a method for optimizing deep artificial neural networks. During the learning process, it reduces the interdependence between nodes by randomly setting some weights or outputs of the hidden layers to zero, thereby regularizing the neural network and reducing its structural risks. For example, in model training, given a vector (1, 2, 3, 4), after inputting this vector into a dropout layer, the dropout layer can randomly convert one of the numbers in the vector (1, 2, 3, 4) to 0. For example, converting 2 to 0 would change the vector to (1, 0, 3, 4).

[0056] Learning rate: Used to control the learning progress of the model. The learning rate guides the model in adjusting network weights using the gradient of the loss function during gradient descent. If the learning rate is too large, the loss function may directly skip the global optimum, resulting in excessive loss. If the learning rate is too small, the loss function changes very slowly, greatly increasing the convergence complexity of the network and making it easy to get trapped in local minima or saddle points.

[0057] Embedded coding, mathematically speaking, represents a correspondence, that is, mapping data in space X to space Y using a function F. This function F is injective, and the mapping result preserves the structure. An injective function means that the mapped data uniquely corresponds to the original data, and preserving the structure means that the order of the original data remains the same. For example, if there are data X1 and X2 before mapping, after mapping we get Y1 corresponding to X1 and Y2 corresponding to X2. If the original data X1 > X2, then correspondingly, the mapped data Y1 > Y2. For words, this means mapping words to another space to facilitate subsequent machine learning and processing.

[0058] Attention weights represent the importance of a piece of data during training or prediction. Importance indicates the magnitude of the influence of input data on output data. Data with high importance corresponds to higher attention weights, while data with low importance corresponds to lower attention weights. The importance of data varies in different scenarios, and training the model to assign attention weights is essentially the process of determining data importance.

[0059] It should be noted that the information (including but not limited to user device information, user personal information, etc.), data (including but not limited to data used for analysis, stored data, displayed data, etc.), and signals involved in the embodiments of this specification are all authorized by the user or fully authorized by all parties, and the collection, use, and processing of related data must comply with the relevant laws, regulations, and standards of the relevant countries and regions. For example, the transaction consultation messages, object information, and interaction behavior information involved in the embodiments of this specification were all obtained with full authorization.

[0060] After introducing the terms used in the embodiments of this specification, the implementation environment of the embodiments of this specification will be described below.

[0061] Figure 1 This is a schematic diagram illustrating the implementation environment of a transaction recommendation method provided in the embodiments of this specification. See also... Figure 1 The implementation environment may include terminal 110 and server 140.

[0062] Terminal 110 is connected to server 140 via a wireless or wired network. Optionally, terminal 110 may be a smartphone, tablet, laptop, desktop computer, smartwatch, etc., but is not limited to these. Terminal 110 has installed and runs applications that support transaction recommendations.

[0063] Server 140 is a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. Server 140 provides background services for applications running on terminal 110. For example, server 140 determines the target transaction to be recommended based on information uploaded by terminal 110 and sends the target transaction to terminal 110 to implement transaction recommendation. In some embodiments, server 140 maintains a tag database and a material database. The tag database stores multiple transaction tags, and the material database stores the materials used when recommending transactions.

[0064] Those skilled in the art will understand that the number of terminals described above can be more or less. For example, there may be only one terminal, or there may be dozens or hundreds of terminals, or even more, in which case other terminals may also be included in the above implementation environment. This specification does not limit the number of terminals or the type of devices in the embodiments.

[0065] After introducing the implementation environment of the embodiments of this specification, the application scenarios of the embodiments of this specification will be described below in conjunction with the above implementation environment. In the following description, the terminal is the terminal 110 in the above implementation environment, and the server is the server 140 in the above implementation environment.

[0066] The technical solutions provided in the embodiments of this specification can be applied to scenarios that recommend various transactions. For example, the technical solutions provided in the embodiments of this specification can be applied to scenarios that recommend insurance products, financial products, and application functions.

[0067] Taking the technical solution provided in the embodiments of this specification as an example in the scenario of recommending insurance products, the transaction is insurance. The server obtains the insurance consultation message entered by the target object on the terminal. The insurance consultation message is used to inquire about information related to the insurance transaction. Based on the insurance consultation message, multiple historical insurance consultation messages of the target object, and the object characteristics of the target object, the server determines the action to be performed on the insurance consultation message. Here, the insurance consultation message is the latest message sent by the target object, and the multiple historical insurance consultation messages are messages sent before this insurance consultation message. The actions include replying to the insurance consultation message and recommending insurance. Replying to the insurance consultation message is also a confirmation of the reply message to the insurance consultation message, thereby guiding the target object to continue sending messages to realize the session; recommending insurance is recommending insurance to the target object. When the determined action is to recommend insurance, the target insurance is determined from multiple candidate insurances and recommended to the target object, thereby completing the insurance recommendation. The technical solution provided in the embodiments of this specification utilizes the target object's insurance consultation message and the target object's object characteristics to perform insurance recommendation, which improves the accuracy of insurance recommendation and thus improves the effect of insurance recommendation.

[0068] It should be noted that the above description is based on the application of the technical solution provided in the embodiments of this specification to the scenario of recommending insurance products. In other possible implementations, the technical solution provided in the embodiments of this specification can also be applied to the scenario of recommending other types of transactions. The embodiments of this specification do not limit the type of transaction.

[0069] After introducing the implementation environment and application scenarios of the embodiments in this specification, the technical solutions provided by the embodiments in this specification are described below. (See also...) Figure 2 Taking the server as the executing entity as an example, the method includes the following steps.

[0070] 202. The server obtains the transaction consultation message for the target object.

[0071] The target object is the entity that wants to participate in the transaction; it is also referred to as the target user, target user account, or target user account, etc. Transaction consultation messages are used to inquire about transactions. In the case of insurance transactions, these messages are also called insurance consultation messages, used to inquire about insurance details, such as the type of insurance, coverage, premiums, or deductibles.

[0072] 204. Based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object, the server determines the action corresponding to the transaction consultation message. The action includes replying to the transaction consultation message and recommending a transaction.

[0073] Among these, several historical transaction consultation messages were sent by the target object prior to this particular transaction consultation message. The target object's characteristics include its interactive behavior characteristics and object profile; of course, both the interactive behavior characteristics and the object profile are used with the target object's consent.

[0074] 206. In the case of a recommended transaction, the server determines the target transaction from multiple candidate transactions based on the transaction consultation message. These multiple candidate transactions are determined based on the multiple historical transaction consultation messages.

[0075] Among them, multiple candidate transactions are multiple transactions recalled by the server. These multiple candidate transactions are determined based on multiple historical transaction consultation messages, and these multiple candidate transactions are transactions that the target object may need.

[0076] 208. The server recommends the target transaction to the target object.

[0077] The target transaction is the transaction that the server determines meets the requirements of the target object.

[0078] The technical solution provided in the embodiments of this specification obtains transaction consultation messages for a target object. Based on these messages, multiple historical transaction consultation messages, and the object characteristics of the target object, the corresponding action is determined. The transaction consultation message and the historical transaction consultation messages reflect the latest needs of the target object, and the object characteristics reflect the object's features. Therefore, the determined action aligns with the target object's personalized needs. When the action is a recommended transaction, the transaction consultation message is used to filter and determine the target transaction from multiple candidate transactions. This target transaction is then recommended to the target object, as it better meets the target object's needs.

[0079] Steps 202-208 above are a brief introduction to the technical solutions provided in the embodiments of this specification. The technical solutions provided in the embodiments of this specification will be explained more clearly below with reference to some examples. See also... Figure 3 The method includes the following steps.

[0080] 302. The server obtains transaction consultation messages for the target object.

[0081] The target object is the entity that wants to participate in the transaction; it is also referred to as the target user, target user account, or target user account, etc. Transaction consultation messages are used to inquire about the transaction. In the case of insurance, these messages are also called insurance consultation messages, used to inquire about insurance details such as the type of insurance, coverage, premium, or deductible. In some embodiments, the insurance consultation message also includes the target object's transaction participation information. This information is required for the target object to participate in the transaction. In the case of insurance, this information includes the policyholder's identity information, such as age, gender, and medical history, as well as the type of insurance the target object desires, such as medical insurance, accident insurance, or travel insurance. Of course, the server needs the user's consent before obtaining the target object's transaction consultation message.

[0082] It should be noted that the transaction consultation message is a message sent by the target object in the current session. The current session is a single session initiated by the target object's consultation transaction. The two parties in the current session are the target object and the chatbot on the server. The chatbot can automatically generate a reply message based on the message sent by the target object.

[0083] In one possible implementation, the terminal sends a transaction consultation message for the target object to the server, the server obtains the transaction consultation message for the target object, and the terminal is the terminal used by the target object.

[0084] In this implementation, the terminal provides a way for the target object to send transaction consultation messages. The target object can use the terminal to quickly send the transaction consultation messages to the server, and the server will perform subsequent processing based on the transaction consultation messages.

[0085] For example, the terminal displays a session interface. In response to a message input operation by the target object on this session interface, the terminal sends a transaction query message corresponding to the message input operation to the server. The server retrieves the transaction query message for the target object. For example, see... Figure 4 The terminal displays a session interface 400, which includes a message input area 401, a message sending control 402, and a message display area 403. The message input area 401 is used to input transaction inquiry messages, the message sending control 402 is used to send the transaction inquiry message to the server, and the message display area 403 is used to display the sent transaction inquiry message and the received reply message. When a transaction inquiry message is input in the message input area 401, in response to a click operation on the message sending control 402, the terminal sends the transaction inquiry message to the server, and the server obtains the transaction inquiry message for the target object. Accordingly, the transaction inquiry message is displayed in the message display area 403. In some embodiments, this session interface is also referred to as an interactive interface.

[0086] In the case of insurance transactions, taking customer interaction as an example, the interface mainly collects user inquiries and questions. For example, if a user asks, "Can I buy insurance if I have diabetes?", the interface will collect the user's question and provide it to downstream services for processing.

[0087] 304. In response to the transaction inquiry message, the server obtains the object characteristics of the target object.

[0088] In one possible implementation, in response to the transaction inquiry message, the server obtains the object information of the target object and its interaction behavior information on the session interface. Based on the object information and the interaction behavior information, the server determines the object characteristics of the target object.

[0089] The session interface is where the target object inputs its inquiry message. Interactions on this interface include clicking, hovering, and swiping. This interaction information records the target object's actions, such as clicking content, hovering over displayed content, and swiping. The target object's information includes its identity and the transactions it has participated in or inquired about. The identity information includes age, gender, and occupation. In the case of insurance, the transactions the target object has participated in or inquired about refer to insurance policies it has purchased or consulted about. It should be noted that both the target object's information and interaction information are obtained with the target object's authorization. In some embodiments, the target object's information is also referred to as its object profile.

[0090] In this implementation, the server can determine the object characteristics of the target object based on the object information of the target object and the interactive behavior information of the target object on the session interface. Since the object information of the target object can reflect the target object's attributes and historical needs for transactions, and the interactive behavior information of the target object on the session interface can reflect the target object's current needs for transactions, the obtained object characteristics can comprehensively reflect the target object's attributes, historical needs for transactions, and current needs for transactions, and thus can comprehensively reflect the target object's needs for transactions.

[0091] For example, in response to the transaction inquiry message, the server retrieves the object information of the target object from the object information database and the interaction behavior information of the target object on the session interface from the terminal. The object information database stores object information of multiple objects, all of which are obtained after authorization. The server merges the object information of the target object and the interaction behavior information to obtain the object features of the target object. For example, in response to the transaction inquiry message, the server queries the object information database based on the identifier of the target object to obtain the object information of the target object. The server retrieves the interaction behavior information of the target object on the session interface uploaded by the terminal. It should be noted that the order in which the server retrieves the object information and the interaction behavior information of the target object can be executed sequentially or simultaneously; this embodiment does not limit this. The server extracts features from the interaction behavior information of the target object on the session interface to obtain the interaction behavior features of the target object. The server concatenates the object information and the interaction behavior features of the target object to obtain the object features of the target object.

[0092] In one possible implementation, in response to the transaction query message, the server obtains the object information of the target object. Based on the object information of the target object and multiple transaction query messages of the target object, the server determines the object characteristics of the target object, wherein the multiple transaction query messages include the current transaction query message and multiple historical transaction query messages.

[0093] Among them, many of the historical transaction consultation messages were sent by the target object before this transaction consultation message.

[0094] In this implementation, the server can determine the object characteristics of the target object based on the object information of the target object and multiple transaction consultation messages. Since the object information of the target object can reflect the target object's attributes and historical needs for transactions, and the multiple transaction consultation messages can reflect the target object's current needs for transactions, the obtained object characteristics can comprehensively reflect the target object's attributes, historical needs for transactions, and current needs for transactions, and thus comprehensively reflect the target object's needs for transactions.

[0095] For example, in response to the transaction inquiry message, the server retrieves the object information of the target object from the object information database. This database stores object information for multiple objects, all of which have been accessed and used with authorization. The server also retrieves multiple transaction inquiry messages for the target object from the message database, which also stores such messages, again accessed and used with authorization. The server then merges the target object's object information with these multiple transaction inquiry messages to obtain the object characteristics of the target object. For instance, in response to the transaction inquiry message, the server queries the object information database based on the target object's identifier to obtain its object information. The server then queries the message database based on the target object's identifier to obtain multiple transaction inquiry messages for the target object. The server performs semantic recognition on these multiple transaction inquiry messages to obtain their semantic features. Finally, the server concatenates these semantic features with the target object's object information to obtain the object characteristics of the target object. The server can employ a semantic recognition model to perform semantic recognition on multiple transaction consultation messages, thereby obtaining the semantic features of the multiple transaction consultation messages. This specification does not limit the type and structure of the semantic recognition model in the embodiments. For example, the semantic recognition model may be a sequence coding model, which can output the semantic features of the multiple transaction consultation messages.

[0096] In one possible implementation, in response to the transaction query message, the server obtains the interaction behavior information of the target object on the session interface. Based on the interaction behavior information and multiple transaction query messages of the target object, the server determines the object characteristics of the target object, including the current transaction query message and multiple historical transaction query messages.

[0097] In this implementation, the server can determine the object characteristics of the target object based on multiple transaction consultation messages of the target object and the interactive behavior information of the target object on the session interface. Since the interactive behavior information of the target object on the session interface can reflect the target object's current needs for the transaction, and the multiple transaction consultation messages can reflect the target object's current needs for the transaction, the obtained object characteristics can more accurately reflect the target object's current needs for the transaction.

[0098] For example, in response to the transaction inquiry message, the server retrieves multiple transaction inquiry messages for the target object from the message database. This database stores transaction inquiry messages for multiple objects, and these messages are retrieved and used only after authorization. The server then retrieves the target object's interactive behavior information on the session interface from the terminal. The server merges the multiple transaction inquiry messages for the target object with this interactive behavior information to obtain the object's characteristics. Alternatively, in response to the transaction inquiry message, the server retrieves multiple transaction inquiry messages for the target object from the message database. This database stores transaction inquiry messages for multiple objects, and these messages are retrieved and used only after authorization. The server retrieves the target object's interactive behavior information uploaded by the terminal on the session interface. The server performs semantic recognition on these multiple transaction inquiry messages to obtain their semantic features. The server extracts features from the target object's interactive behavior information on the session interface to obtain its interactive behavior features. Finally, the server concatenates the semantic features of the multiple transaction inquiry messages with the interactive behavior features to obtain the target object's characteristics.

[0099] It should be noted that the server can use any of the above methods to obtain the object characteristics of the target object, and the embodiments in this specification do not limit this.

[0100] 306. The server determines the status of the target object based on multiple historical transaction consultation messages of the target object and the object characteristics of the target object.

[0101] Among these, multiple historical transaction consultation messages are messages sent by the target object prior to this current consultation message; these multiple historical transaction consultation messages are also referred to as the target object's historical sessions. In some embodiments, the target object's object characteristics include the target object's interaction behavior characteristics and object profile; of course, both the interaction behavior characteristics and the object profile are used with the target object's consent. In some embodiments, determining the state of the target object is also referred to as performing state modeling on the target object.

[0102] In one possible implementation, the server inputs multiple historical transaction consultation messages of the target object and the object features of the target object into a state determination model. The server uses the state determination model to extract features from the multiple historical transaction consultation messages to obtain contextual features of the multiple historical transaction consultation messages. The server then uses the state determination model to fuse the contextual features of the multiple historical transaction consultation messages with the object features of the target object, and outputs the state of the target object.

[0103] The state of the target object is also called the hidden state of the target object, and subsequent actions can be predicted based on the hidden state.

[0104] In this implementation, the server can extract features from multiple historical transaction consultation messages using a state determination model to obtain contextual features of these messages. The server then fuses these contextual features with the object features of the target object using the state determination model to obtain the state of the target object. Since the contextual features reflect the target object's needs, and the object features reflect its attributes, the obtained state comprehensively reflects both the target object's needs and attributes, facilitating subsequent decision-making based on the target object's state.

[0105] To provide a clearer explanation of the above embodiments, the following description will be divided into several parts.

[0106] The first part involves the server using this state to determine the model and extracting features from the multiple historical transaction consultation messages to obtain the contextual features of these messages.

[0107] In one possible implementation, the server uses the state determination model to perform time-series encoding on the multiple historical transaction consultation messages to obtain the contextual features of the multiple historical transaction consultation messages. Accordingly, the state determination model includes a time-series encoding sub-model, through which the multiple historical transaction consultation messages can be time-series encoded.

[0108] In this implementation, the temporal characteristics of multiple historical transaction consultation messages are learned through temporal coding, thereby obtaining the contextual features of these multiple historical transaction consultation messages. Since temporal coding can make full use of the temporal continuity of multiple historical transaction consultation messages, the obtained contextual features can more accurately represent the semantics of these multiple historical transaction consultation messages.

[0109] For example, for the first historical transaction consultation message among multiple historical transaction consultation messages, the server uses the temporal coding sub-model of the state determination model and employs a gating mechanism to encode the first historical transaction consultation message, obtaining its first hidden state. This first historical transaction consultation message is the earliest published historical transaction consultation message among the multiple historical transaction consultation messages. For the second historical transaction consultation message among the multiple historical transaction consultation messages, the server uses the temporal coding sub-model of the state determination model and employs a gating mechanism to encode both the second historical transaction consultation message and its first hidden state, obtaining its second hidden state. This process continues until the temporal coding of all the historical transaction consultation messages is completed, and the final hidden state obtained is the context feature. In some embodiments, the temporal coding sub-model is a Long Short Term Memory Network (LSTM).

[0110] In one possible implementation, the server uses this state determination model to encode the multiple historical transaction consultation messages based on an attention mechanism, thereby obtaining the contextual features of the multiple historical transaction consultation messages. Accordingly, the state determination model includes an attention encoding sub-model, which enables the encoding of the multiple historical transaction consultation messages based on an attention mechanism.

[0111] In this implementation, the attention mechanism can be used to fully model the multiple historical transaction consultation messages. During the modeling process, more attention is paid to the important historical transaction consultation messages, so that the obtained context features can more accurately reflect the semantics of the multiple historical transaction consultation messages.

[0112] For example, the server uses the attention encoding sub-model of the state determination model to embed and encode multiple historical transaction consultation messages, obtaining the embedding vectors of each message. The server then uses this attention encoding sub-model to map the multiple embedding vectors of these historical transaction consultation messages, obtaining the query matrix, key matrix, and value matrix for each message. Based on the query and key matrices, the server uses the attention encoding sub-model to obtain the attention weights for each historical transaction consultation message. Finally, the server uses the attention weights to weight and fuse the value matrices of each historical transaction consultation message, obtaining the contextual features of the multiple historical transaction consultation messages.

[0113] In one possible implementation, the server uses the state determination model to embed and encode the multiple historical transaction consultation messages, obtaining the embedding vectors of each message. The server then uses the state model to convolve multiple embedding vectors from the historical transaction consultation messages to obtain the contextual features of the messages. Accordingly, the state determination model includes a convolutional sub-model, which enables the convolution of the multiple historical transaction consultation messages.

[0114] In this implementation, convolution can be used to learn the local features between different historical transaction consultation messages, and the context features formed by the local features can more accurately reflect the semantics of the multiple historical transaction consultation messages.

[0115] For example, the server determines the attention encoding sub-model of the model based on this state, performs embedding encoding on the multiple historical transaction consultation messages, and obtains the embedding vector of each historical transaction consultation message. The server concatenates the multiple embedding vectors of the multiple historical transaction consultation messages into an embedding vector matrix, and slides at least one convolution kernel on the embedding vector matrix. During the sliding process, it performs convolution operations with the covered positions, and outputs the context features of the multiple historical transaction consultation messages.

[0116] It should be noted that the server can determine the model through this state and use any of the above methods to output the contextual features of the multiple historical transaction consultation messages. This specification does not limit this aspect in the embodiments.

[0117] The second part involves the server using this state to determine the model, fusing the contextual features of the multiple historical transaction consultation messages with the object features of the target object, and outputting the state of the target object.

[0118] In one possible implementation, the server uses the state determination model to perform a full connection and normalization on the context features of the multiple historical transaction consultation messages and the object features of the target object, and outputs the state of the target object.

[0119] Normalization can be performed using any of the Sigmoid (S-shaped growth) function, ReLU (linear rectification) function, or Softmax (soft maximization) function, but this specification does not limit the specific implementation of the embodiments.

[0120] In this implementation, the server can determine the model through the state, and fully integrate the context features and the object features using a fully connected and normalized approach, so that the resulting state can accurately represent the target object.

[0121] For example, the server concatenates the contextual features of multiple historical transaction consultation messages with the object features of the target object to form a state determination matrix. Using this state determination model, the server performs at least one full connection and normalization on the state determination matrix to output the state of the target object.

[0122] In one possible implementation, the server uses the state determination model to add the contextual features of the multiple historical transaction consultation messages to the object features of the target object, and outputs the state of the target object.

[0123] In this implementation, the server can quickly fuse the context features and the object features by adding them together, thus improving the efficiency of feature fusion.

[0124] For example, the server uses this state to determine the model, unifying the contextual features of multiple historical transaction consultation messages and the object features of the target object into a target dimension. The server then uses this state to determine the model, adding the contextual features of the target dimension to the object features of that target dimension, and outputting the state of the target object.

[0125] 308. The server determines the action corresponding to the transaction consultation message based on the state of the target object and the transaction consultation message.

[0126] The actions include replying to the transaction inquiry message and recommending a transaction. Replying to the transaction inquiry message means generating a reply message to continue the conversation with the target object, thereby guiding the target object to send more transaction inquiry messages. Recommending a transaction means recommending a transaction to the target object; in the case of insurance, this means recommending insurance to the target object. In some embodiments, replying to the transaction inquiry message is also referred to as "tagging an inquiry" with the user, and recommending a transaction is also referred to as "recommending a product" to the user.

[0127] In one possible implementation, the server inputs the state of the target object and the transaction consultation message into an action determination model. Using this action determination model, the server maps the state of the target object to the transaction consultation message and outputs the action corresponding to the transaction consultation message.

[0128] In this implementation, the server can input the state of the target object and the transaction consultation message into the action determination model, and determine the action corresponding to the transaction consultation message through the action determination model, which has high efficiency in action determination.

[0129] For example, the server uses the action determination model to perform a full join and normalization on the state of the target object and the transaction query message, and outputs the action corresponding to the transaction query message. Alternatively, the server concatenates the state of the target object and the transaction query message into an action determination matrix. The server then uses this action determination model to perform at least one full join and normalization on the action determination matrix, and outputs the action corresponding to the transaction query message.

[0130] In some embodiments, the action determination model is obtained through reinforcement learning training.

[0131] In one possible implementation, the server generates a target graph based on the state of the target object and the transaction query message. The server then processes the target graph using a graph neural network to obtain the action corresponding to the transaction query message.

[0132] In this implementation, the server can process the target graph through a graph neural network to obtain the action corresponding to the transaction consultation message. The graph neural network is easy to train, has a fast processing speed, and is highly efficient in determining the action.

[0133] For example, the server creates multiple nodes based on the state of the target object and the transaction query message. These nodes include object nodes, state nodes, and message nodes for the target object. The server adds connections between these nodes to obtain the target graph. The server then performs graph convolution on the target graph using a graph neural network to obtain the action corresponding to the transaction query message.

[0134] Optionally, after step 308, the server may execute steps 310-312 or steps 314-316, which are not limited in this embodiment of the specification.

[0135] 310. In the case of a recommended transaction, the server determines the target transaction from multiple candidate transactions based on the transaction consultation message. These multiple candidate transactions are determined based on the multiple historical transaction consultation messages.

[0136] The multiple candidate transactions are those recalled by the server. These candidate transactions are determined based on the multiple historical transaction consultation messages, and they represent transactions that the target object might need. In the case of transactions as insurance, these multiple candidate transactions represent the insurance that the target object might need.

[0137] In one possible implementation, the server performs interest extraction on the transaction consultation message to obtain the points of interest within the message. Based on these points of interest, the server determines the target transaction from among the multiple candidate transactions.

[0138] Points of interest (POIs) reflect the target audience's interest in a transaction; in some embodiments, POIs are also referred to as keywords. In the case of an insurance transaction, the POI can reflect the target audience's interest in insurance.

[0139] In this implementation, the server can obtain points of interest from transaction consultation messages and determine the target transaction from multiple candidate transactions based on the points of interest. The target transaction is the transaction that the target object may need.

[0140] To provide a clearer explanation of the above embodiments, the following description will be divided into two parts.

[0141] Part 1: The server extracts the points of interest from the transaction consultation message.

[0142] In one possible implementation, the server performs slot extraction and standardization on the transaction consultation message to obtain the points of interest in the transaction consultation message.

[0143] In this context, a slot is a clearly defined attribute of an entity. For example, in a ride-hailing scenario, departure location A is a slot, destination B is a slot, and departure time C is a slot, corresponding to the three slots of "departure location," "destination," and "departure time," respectively. A slot is a way of filling slots. Standardization is a method of converting vocabulary into standard terms. For instance, if there is a non-standard term "lower abdomen," standardization can yield the standard term "lower abdomen." By standardizing vocabulary, different expressions of the same thing can be unified, facilitating subsequent processing.

[0144] For example, the server inputs the transaction inquiry message into a slot extraction model. This model annotates multiple characters in the message, obtaining labels that represent the type of each character. The server then determines the slots in the message based on the annotation results. Next, the server inputs these slots into a lexical normalization model, which determines the standard word corresponding to each slot. This standard word represents the point of interest in the message. Note that the slot extraction model and the lexical normalization model can be arbitrary structures; this specification does not limit these specific methods.

[0145] The second part involves the server identifying the target transaction from among multiple candidate transactions based on the points of interest in the transaction consultation message.

[0146] In one possible implementation, the server performs tag mapping on the points of interest in the transaction consultation message to obtain the transaction tags corresponding to the points of interest in the transaction consultation message. The server uses the transaction tags to filter among multiple candidate transactions to obtain the target transaction, which has the transaction tag.

[0147] In this implementation, the server can determine transaction tags based on points of interest, and then filter transactions based on these tags, resulting in a high accuracy in identifying target transactions.

[0148] For example, the server matches points of interest in a tag database to obtain the transaction tag corresponding to the point of interest. The semantic similarity between the transaction tag and the point of interest meets the similarity condition. The server uses this transaction tag to perform tag matching among multiple candidate transactions to obtain the target transaction, which has the transaction tag. The tag database stores multiple preset transaction tags; matching within this database means determining the transaction tag with the highest similarity among these preset transaction tags. In some embodiments, semantic similarity meeting the similarity condition means that the semantic similarity is greater than or equal to a semantic similarity threshold. This semantic similarity threshold is set by a technician according to actual conditions, and this specification does not limit it in this embodiment.

[0149] Taking insurance as an example, if the inquiry message is "I want to buy a medical product under 1000 yuan", the server extracts the interest from the inquiry message and gets two interest points: "under 1000 yuan" and "medical product". The server standardizes the interest point "under 1000 yuan" into the label "insurance cost" and the interest point "medical product" into the label "medical insurance".

[0150] The above embodiments involve concepts such as candidate transactions and preset transaction tags. In order to explain the above embodiments more clearly, the method for determining the candidate transactions and preset transaction tags will be described below.

[0151] First, the method for determining the candidate transaction will be explained.

[0152] In one possible implementation, the server performs interest extraction on the multiple historical transaction consultation messages to obtain multiple points of interest within the messages. Based on these points of interest, the server then retrieves multiple candidate transactions.

[0153] For example, the server extracts and standardizes slots from the multiple transaction consultation messages to obtain multiple points of interest (POIs). The server then maps these POIs to tags, obtaining multiple transaction tags corresponding to them. Finally, the server matches these transaction tags with multiple preset transactions to obtain multiple candidate transactions.

[0154] The process by which the server matches the multiple transaction tags with multiple preset transactions to obtain the multiple candidate transactions is a tag-based recall method. That is, the server matches the multiple transaction tags with the tags of multiple preset transactions, and the preset transactions that are successfully matched are the candidate transactions.

[0155] In some embodiments, the server performs multi-path recall based on multiple points of interest in the multiple historical transaction consultation messages to obtain multiple candidate transactions. Multi-path recall refers to recalling transactions in different ways; for example, the server can determine at least one of the following based on the multiple points of interest: transaction type, transaction attribute, and transaction individual. The server performs recall based on at least one of the following: transaction type, transaction attribute, and transaction individual to obtain multiple candidate transactions. In the case of insurance transactions, that is, the server recalls eligible insurance products through different forms such as "insurance type recall," "product recall," and "attribute recall."

[0156] In some embodiments, in addition to recalling the multiple candidate transactions, the server can also recall multiple reference transaction tags corresponding to the multiple candidate transactions, and the multiple reference transaction tags are all transaction tags associated with the multiple candidate transactions.

[0157] It should be noted that since the multiple historical transaction consultation messages are sent by the target object in multiple batches, the server can perform transaction recall based on each historical transaction consultation message after receiving it. By continuously recalling these messages, multiple candidate transactions can be obtained.

[0158] The method for determining the preset transaction label is then explained.

[0159] In one possible implementation, the server structures the introductory text of multiple preset transactions to obtain the multiple preset transaction tags.

[0160] Here, "structuring" refers to document structuring technology, which allows the extraction of multiple pre-defined transaction tags. The introductory text of a pre-defined transaction refers to its relevant terms and conditions, outlining the rights and obligations of participating in the transaction. In the case of insurance, the introductory text of the pre-defined transaction is also the insurance terms and conditions. In some embodiments, the server uses document structuring technology to extract a series of tags from the insurance terms and conditions, constructing an insurance product tag map. The tag database includes insurance-related tags such as renewal rules and insurance premiums.

[0161] 312. The server recommends the target transaction to the target object.

[0162] The target transaction is the transaction that the server determines meets the requirements of the target object.

[0163] In one possible implementation, the server sends the interaction card for the target transaction to the terminal used by the target object.

[0164] Interactive cards, in this context, refer to cards that present tag content and related options in an interactive format. In some embodiments, the interactive card includes a text display area and interactive controls. The text display area displays text related to the target transaction. For example, if the target transaction is insurance, the text may include at least one of the following: a recommendation reason, tag text, and investor education content. The recommendation reason explains the selling points and highlights of the insurance product; the tag text provides explanations of the corresponding tags in the tag database; and the investor education content further explains the tag text, aiming to enhance the user's understanding of insurance. The interactive controls provide the ability for the target object to interact with the interactive card. In the case of insurance, the interactive controls can be used for purchasing insurance.

[0165] For example, the server queries the text database based on the target transaction and retrieves the text corresponding to the target transaction. The server then fills the interactive card template with the text corresponding to the target transaction to obtain the interactive card for that target transaction. The server then sends the interactive card for the target transaction to the terminal used by the target user.

[0166] In one possible implementation, the server sends media resources of the target transaction to the terminal used by the target object, the media resources being used to introduce the target transaction.

[0167] Media resources are a form of copywriting, including videos and images. Through images and videos, the content in the copywriting library is presented to the target audience.

[0168] For example, the server queries the document database based on the target transaction to obtain the document corresponding to the target transaction. Based on the document corresponding to the target transaction, the server generates the media resource corresponding to the target transaction. The server then sends the media resource corresponding to the target transaction to the terminal used by the target user.

[0169] In one possible implementation, the server sends a virtual object control instruction to the terminal using the target object, which instructs the terminal to control the virtual object to introduce the target transaction.

[0170] Virtual objects, also known as virtual people or digital people, combine the image of virtual objects with the ability of text-to-speech (TTS) to explain relevant information to the target audience through a combination of visual and auditory means, which can enhance users' willingness to interact.

[0171] For example, the server sends a virtual object control instruction to the terminal using the target object. This virtual object control instruction carries the text corresponding to the target transaction. The terminal receives the virtual object control instruction and, based on this instruction, controls the virtual object to use the text corresponding to the target transaction to introduce the target transaction.

[0172] 314. If the action is to reply to the transaction inquiry message, the server determines the transaction tag corresponding to the transaction inquiry message.

[0173] The transaction tag corresponding to the transaction consultation message can reflect the intent of the transaction consultation message.

[0174] In one possible implementation, when the action is to reply to the transaction inquiry message, the server performs semantic recognition on the transaction inquiry message to obtain its semantic features. Based on these semantic features, the server performs semantic matching in a tag database to obtain the transaction tag corresponding to the transaction inquiry message.

[0175] For example, when the action is to reply to the transaction inquiry message, the server inputs the transaction inquiry message into a semantic recognition model, performs semantic recognition on the message, and obtains its semantic features. The server determines the similarity between the semantic features of the message and the semantic features of multiple preset transaction tags in a tag database, and identifies the preset transaction tags whose semantic similarity meets the similarity criteria as the corresponding transaction tags for the message. The semantic recognition model can be of any structure, and this embodiment does not limit its implementation.

[0176] 316. The server determines the response message to the inquiry message based on the transaction tag.

[0177] In one possible implementation, the server performs a message query among multiple candidate response messages based on the transaction tag to obtain at least one response message corresponding to the transaction tag. These multiple candidate response messages are pre-configured by the technical personnel. For example, if the transaction tag is "Medical Green Channel," the server further determines what content to display to the target audience. The content presented to the target audience might be, "In terms of product services, would you need more coverage including Medical Green Channel coverage?"

[0178] In one possible implementation, the server inputs the transaction tag into a message generation model, which then generates text based on the transaction tag to obtain a response message to the transaction inquiry message. The message generation model can be any structured text generation model, and this specification does not limit its implementation.

[0179] In some embodiments, step 316 is performed by a chatbot on the server.

[0180] In the case of insurance transactions, the technical solutions provided in the embodiments of this specification can be used by insurance agents when selling insurance products to customers, or customers can choose insurance products themselves through a chatbot. For users whose intention to "purchase insurance" is recognized by the system, the technical solutions provided in the embodiments of this specification determine questions that the user may be interested in based on the user's profile information and historical conversations. Feedback is collected based on the user's options, and the system then decides on the next steps for interaction with the user or the recommended insurance products.

[0181] The following will combine Figure 5 The technical solutions provided in the embodiments of this specification will be described.

[0182] See Figure 5Taking insurance as an example, the technical solution provided in this specification can be implemented by an interactive insurance recommendation system deployed on a server. This interactive insurance recommendation system comprises three parts: an interactive interface 501, an insurance recommendation decision center 502, and an insurance database 503. The interactive interface 501 is the same as the conversation interface in step 301 above. The interactive interface 501 includes an insurance agent interface and a target object interface. The insurance agent interface is for insurance agents, and the target object interface is for insurance customers. For insurance agents, they can interact with the target object through questions provided on the interactive interface 501; for insurance customers, they can select suitable insurance products through the interactive interface 501. The insurance recommendation decision center 502 includes an interest collection unit 5021, an interest recall unit 5022, a dialogue decision unit 5023, and a recommendation unit 5024. The interest collection unit 5021 processes the information obtained from the interactive interface 501 to obtain the target object's points of interest. The interest collection unit 5021 includes an intent recognition subunit, an interest extraction subunit, and a behavior and profiling subunit. The intent recognition subunit analyzes the target object's questions using text classification techniques in natural language processing to identify the target object's intent to purchase insurance, thus executing the relevant steps in step 310. The interest extraction subunit extracts the target object's interest points using slot extraction and standardization techniques in natural language processing and maps them to tag names in a tag database, also executing the relevant steps in step 310. The behavior and profiling subunit collects the target object's behaviors on the interactive interface, such as "clicks" and "stays," as well as historical target object profiles. This behavior and profiling subunit serves the dialogue decision unit 5023, executing step 304. The interest recall unit 5022 recalls insurance based on the interest points determined by the interest collection unit 5021. For example, it can recall eligible insurance products and associated tags through different forms such as "insurance type recall," "product recall," and "attribute recall." The dialogue decision unit 5023 employs deep learning technology to model the target object's historical dialogues and profile, and uses algorithms such as reinforcement learning / graph neural networks to determine appropriate actions for interaction with the target object. Actions include two main categories: "tag inquiry" and "product recommendation." This dialogue decision unit 5023 comprises a dialogue memory subunit, a state modeling subunit, and an action decision subunit. The dialogue memory subunit stores the target object's historical dialogues, i.e., records multiple historical transaction inquiry messages. The state modeling subunit models the target object's state, performing step 306. The action decision subunit determines the action corresponding to the transaction inquiry message, performing step 308.The recommendation unit 5024 includes a session determination subunit and a product recommendation subunit. The session determination subunit is used to determine the response message corresponding to the transaction inquiry message and to perform steps 314 and 316 above. The product recommendation subunit is used to recommend insurance products and to perform steps 310 and 312 above.

[0183] The insurance database 503 includes a tag database 5031 and a material database 5032. The tag database 5031 includes an insurance type database, a product database, and an attribute database. The insurance type database stores insurance type tags, such as medical insurance and critical illness insurance; the product database stores product tags, such as hospitalization insurance and cancer insurance; and the attribute database stores attribute tags, such as renewal rules and insurance fees. The material database 5032 includes a copywriting database and an interaction database. The copywriting database includes recommendation reasons, tag copywriting, and investor education content. The interaction database includes interactive cards, images / videos, and digital humans.

[0184] All the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this specification, and will not be described in detail here.

[0185] The technical solution provided in the embodiments of this specification obtains transaction consultation messages for a target object. Based on these messages, multiple historical transaction consultation messages, and the object characteristics of the target object, the corresponding action is determined. The transaction consultation message and the historical transaction consultation messages reflect the latest needs of the target object, and the object characteristics reflect the object's features. Therefore, the determined action aligns with the target object's personalized needs. When the action is a recommended transaction, the transaction consultation message is used to filter and determine the target transaction from multiple candidate transactions. This target transaction is then recommended to the target object, as it better meets the target object's needs.

[0186] When applied to insurance recommendation scenarios, the technical solutions provided in the embodiments of this specification are beneficial to users, agents, and insurance platforms. For the insurance industry, the technical solutions provided in the embodiments of this specification can increase premium income, improve operational efficiency, enhance user experience, reduce cost risks, and help the insurance industry transform towards "digitalization."

[0187] Figure 6 This is a schematic diagram of the structure of a transaction recommendation device provided in an embodiment of this specification. See also... Figure 6 The device includes: a message acquisition module 601, an action determination module 602, a transaction determination module 603, and a recommendation module 604.

[0188] The message acquisition module 601 is used to acquire transaction consultation messages of the target object.

[0189] The action determination module 602 is used to determine the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object. The action includes replying to the transaction consultation message and recommending a transaction.

[0190] The transaction determination module 603 is used to determine the target transaction from multiple candidate transactions based on the transaction consultation message when the action is a recommended transaction. The multiple candidate transactions are determined based on the multiple historical transaction consultation messages.

[0191] Recommendation module 604 is used to recommend the target transaction to the target object.

[0192] In one possible implementation, the action determination module 602 is used to determine the state of the target object based on multiple historical transaction consultation messages of the target object and the object characteristics of the target object. Based on the state of the target object and the transaction consultation message, the action corresponding to the transaction consultation message is determined.

[0193] In one possible implementation, the action determination module 602 is used to input multiple historical transaction consultation messages of the target object and the object features of the target object into a state determination model. The state determination model extracts features from the multiple historical transaction consultation messages to obtain their contextual features. The state determination model then fuses the contextual features of the multiple historical transaction consultation messages with the object features of the target object to output the state of the target object.

[0194] In one possible implementation, the action determination module 602 is used to input the state of the target object and the transaction consultation message into the action determination model. The action determination model maps the state of the target object and the transaction consultation message, and outputs the action corresponding to the transaction consultation message.

[0195] In one possible implementation, the transaction determination module 603 is used to extract interests from the transaction consultation message to obtain points of interest in the transaction consultation message. Based on the points of interest in the transaction consultation message, the target transaction is determined from the plurality of candidate transactions.

[0196] In one possible implementation, the transaction determination module 603 is used to perform tag mapping on the points of interest in the transaction consultation message to obtain the transaction tags corresponding to the points of interest in the transaction consultation message. The target transaction is obtained by filtering among multiple candidate transactions using these transaction tags, and the target transaction has the transaction tag.

[0197] In one possible implementation, the recommendation module 604 is configured to perform any of the following:

[0198] Send the interaction card for the target transaction to the terminal used by the target object.

[0199] The media resources of the target transaction are sent to the terminal used by the target object to introduce the target transaction.

[0200] Send a virtual object control instruction to the terminal used by the target object. The virtual object control instruction is used to instruct the terminal to control the virtual object to introduce the target transaction.

[0201] In one possible implementation, the device further includes:

[0202] The object feature acquisition module is used to acquire object information and interactive behavior information of the target object on the session interface. Based on the object information and interactive behavior information, the object features of the target object are determined.

[0203] In one possible implementation, the device further includes:

[0204] The recall module is used to extract interests from the multiple historical transaction consultation messages, obtaining multiple points of interest within these messages. Based on these multiple points of interest, a recall process is performed to obtain multiple candidate transactions.

[0205] In one possible implementation, the recall module is used to perform tag mapping on the multiple points of interest to obtain multiple transaction tags corresponding to the multiple points of interest. These multiple transaction tags are then matched with multiple preset transactions to obtain multiple candidate transactions.

[0206] In one possible implementation, the device further includes:

[0207] The structuring module is used to structure the introductory text of multiple preset transactions to obtain the tags for those preset transactions.

[0208] In one possible implementation, the device further includes:

[0209] The intent recognition module is used to identify the intents of the multiple historical transaction consultation messages, thereby obtaining multiple intents corresponding to the multiple historical transaction consultation messages. If any of the multiple intents is the target intent, the module performs interest extraction on the multiple historical transaction consultation messages to obtain multiple points of interest in the multiple historical transaction consultation messages.

[0210] In one possible implementation, the device further includes:

[0211] The message determination module is used to determine the transaction tag corresponding to the transaction inquiry message when the action is to reply to the transaction inquiry message. Based on the transaction tag, the reply message for the transaction inquiry message is determined.

[0212] It should be noted that the transaction recommendation device provided in the above embodiments is only illustrated by the division of the above functional modules when recommending transactions. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the transaction recommendation device and the transaction recommendation method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0213] The technical solution provided in the embodiments of this specification obtains transaction consultation messages for a target object. Based on these messages, multiple historical transaction consultation messages, and the object characteristics of the target object, the corresponding action is determined. The transaction consultation message and the historical transaction consultation messages reflect the latest needs of the target object, and the object characteristics reflect the object's features. Therefore, the determined action aligns with the target object's personalized needs. When the action is a recommended transaction, the transaction consultation message is used to filter and determine the target transaction from multiple candidate transactions. This target transaction is then recommended to the target object, as it better meets the target object's needs.

[0214] This specification provides a computer device for performing the above-described method. This computer device can be implemented as a terminal or a server. The structure of a terminal will be described below:

[0215] Figure 7 This is a schematic diagram of the structure of a terminal provided in an embodiment of this specification. The terminal 700 can be a smartphone, tablet computer, laptop computer, or desktop computer. The terminal 700 may also be referred to as user equipment, portable terminal, laptop terminal, desktop terminal, or other names.

[0216] Typically, terminal 700 includes one or more processors 701 and one or more memories 702.

[0217] Processor 701 may include one or more processing cores, such as a quad-core processor, an octa-core processor, etc. Processor 701 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). Processor 701 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, processor 701 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, processor 701 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0218] Memory 702 may include one or more computer-readable storage media, which may be non-transitory. Memory 702 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the non-transitory computer-readable storage media in memory 702 are used to store at least one computer program, which is executed by processor 701 to implement the transaction recommendation method provided in the method embodiments of this specification.

[0219] In some embodiments, the terminal 700 may also optionally include a peripheral device interface 703 and at least one peripheral device. The processor 701, memory 702, and peripheral device interface 703 can be connected via a bus or signal line. Each peripheral device can be connected to the peripheral device interface 703 via a bus, signal line, or circuit board. Specifically, the peripheral device includes at least one of the following: a radio frequency circuit 704, a display screen 705, a camera assembly 706, an audio circuit 707, and a power supply 708.

[0220] Peripheral device interface 703 can be used to connect at least one I / O (Input / Output) related peripheral device to processor 701 and memory 702. In some embodiments, processor 701, memory 702 and peripheral device interface 703 are integrated on the same chip or circuit board; in some other embodiments, any one or two of processor 701, memory 702 and peripheral device interface 703 can be implemented on separate chips or circuit boards, which is not limited in this embodiment.

[0221] The radio frequency (RF) circuit 704 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The RF circuit 704 communicates with communication networks and other communication devices via electromagnetic signals. The RF circuit 704 converts electrical signals into electromagnetic signals for transmission, or converts received electromagnetic signals back into electrical signals. Optionally, the RF circuit 704 includes: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a user identity module card, etc.

[0222] Display screen 705 is used to display a user interface (UI). This UI may include graphics, text, icons, video, and any combination thereof. When display screen 705 is a touch display screen, it also has the ability to collect touch signals on or above its surface. These touch signals can be input as control signals to processor 701 for processing. In this case, display screen 705 can also be used to provide virtual buttons and / or a virtual keyboard, also known as soft buttons and / or a soft keyboard.

[0223] The camera assembly 706 is used to capture images or videos. Optionally, the camera assembly 706 includes a front-facing camera and a rear-facing camera. Typically, the front-facing camera is located on the front panel of the terminal, and the rear-facing camera is located on the back of the terminal.

[0224] The audio circuit 707 may include a microphone and a speaker. The microphone is used to collect sound waves from the user and the environment, and convert the sound waves into electrical signals that are input to the processor 701 for processing, or input to the radio frequency circuit 704 to realize voice communication.

[0225] The power supply 708 is used to supply power to the various components in the terminal 700. The power supply 708 can be AC ​​power, DC power, a disposable battery, or a rechargeable battery.

[0226] In some embodiments, the terminal 700 further includes one or more sensors 709. The one or more sensors 709 include, but are not limited to: an accelerometer 710, a gyroscope 711, a pressure sensor 712, an optical sensor 713, and a proximity sensor 714.

[0227] Accelerometer 710 can detect the magnitude of acceleration on the three coordinate axes of a coordinate system established with terminal 700.

[0228] The gyroscope sensor 711 can detect the orientation and rotation angle of the terminal 700. The gyroscope sensor 711 can work in conjunction with the accelerometer sensor 710 to collect the user's 3D movements on the terminal 700.

[0229] The pressure sensor 712 can be installed on the side bezel of the terminal 700 and / or on the lower layer of the display screen 705. When the pressure sensor 712 is installed on the side bezel of the terminal 700, it can detect the user's grip signal on the terminal 700, and the processor 701 can perform left / right hand recognition or quick operation based on the grip signal collected by the pressure sensor 712. When the pressure sensor 712 is installed on the lower layer of the display screen 705, the processor 701 can control the operable controls on the UI interface based on the user's pressure operation on the display screen 705.

[0230] An optical sensor 713 is used to collect ambient light intensity. In one embodiment, the processor 701 can control the display brightness of the display screen 705 based on the ambient light intensity collected by the optical sensor 713.

[0231] The proximity sensor 714 is used to detect the distance between the user and the front of the terminal 700.

[0232] Those skilled in the art will understand that Figure 7 The structure shown does not constitute a limitation on terminal 700, and may include more or fewer components than shown, or combine certain components, or use different component arrangements.

[0233] The aforementioned computer equipment can also be implemented as a server. The structure of a server is described below:

[0234] Figure 8This is a schematic diagram of a server structure provided in an embodiment of this specification. The server 800 can vary significantly due to differences in configuration or performance. It may include one or more Central Processing Units (CPUs) 801 and one or more memories 802. The one or more memories 802 store at least one computer program, which is loaded and executed by the one or more processors 801 to implement the methods provided in the various method embodiments described above. Of course, the server 800 may also have wired or wireless network interfaces, a keyboard, and input / output interfaces for input and output. The server 800 may also include other components for implementing device functions, which will not be elaborated upon here.

[0235] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including a computer program that can be executed by a processor to perform the transaction recommendation method in the above embodiments. For example, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage device, etc.

[0236] In an exemplary embodiment, a computer program product or computer program is also provided, which includes program code stored in a computer-readable storage medium. A processor of a computer device reads the program code from the computer-readable storage medium and executes the program code, causing the computer device to perform the transaction recommendation method described above.

[0237] In some embodiments, the computer program described in this specification may be deployed and executed on a single computer device, or on multiple computer devices located in one location, or on multiple computer devices distributed across multiple locations and interconnected via a communication network. These multiple computer devices distributed across multiple locations and interconnected via a communication network may constitute a blockchain system.

[0238] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0239] The above are merely optional embodiments of this specification and are not intended to limit this specification. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this specification shall be included within the scope of protection of this specification.

Claims

1. A transaction recommendation method, the method comprising: Retrieve transaction consultation messages from the target object; Based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object, the action corresponding to the transaction consultation message is determined, and the action includes replying to the transaction consultation message and recommending a transaction. In the case of a recommended transaction, a target transaction is determined from multiple candidate transactions based on the transaction consultation message. The multiple candidate transactions are obtained by continuous recall. The specific method of continuous recall is to recall transactions based on each historical transaction consultation message after receiving it. Recommend the target transaction to the target object; The step of determining the target transaction from multiple candidate transactions based on the transaction consultation message includes: Interest extraction is performed on the transaction consultation message to obtain the points of interest in the transaction consultation message; Based on the points of interest in the transaction consultation message, the target transaction is determined from the plurality of candidate transactions; The step of determining the target transaction from the plurality of candidate transactions based on the points of interest in the transaction consultation message includes: The points of interest in the transaction consultation message are labeled and mapped to obtain the transaction labels corresponding to the points of interest in the transaction consultation message; The target transaction is obtained by filtering among the multiple candidate transactions using the transaction tag, and the target transaction has the transaction tag. In the case where the action is a recommended transaction, before determining the target transaction from multiple candidate transactions based on the transaction consultation message, the method further includes: Interest extraction is performed on the multiple historical transaction consultation messages to obtain multiple points of interest in the multiple historical transaction consultation messages; Based on multiple points of interest in the multiple historical transaction consultation messages, multiple candidate transactions are obtained by multi-path recall, where multi-path recall refers to recalling in different ways; The process of multi-path recall based on multiple points of interest in the multiple historical transaction consultation messages to obtain the multiple candidate transactions includes: The multiple points of interest are labeled to obtain multiple transaction labels corresponding to the multiple points of interest. The transaction label includes at least one of transaction type, transaction attribute and transaction individual. The multiple transaction tags are matched with multiple preset transactions to obtain the multiple candidate transactions.

2. The method according to claim 1, wherein determining the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object includes: The state of the target object is determined based on multiple historical transaction consultation messages of the target object and the object characteristics of the target object; Based on the state of the target object and the transaction consultation message, determine the action corresponding to the transaction consultation message.

3. The method according to claim 2, wherein determining the state of the target object based on multiple historical transaction consultation messages of the target object and the object characteristics of the target object includes: The target object's multiple historical transaction consultation messages and the target object's object characteristics are input into the state determination model; By using the state determination model, feature extraction is performed on the multiple historical transaction consultation messages to obtain the context features of the multiple historical transaction consultation messages; The state determination model integrates the contextual features of the multiple historical transaction consultation messages with the object features of the target object to output the state of the target object.

4. The method according to claim 2, wherein determining the action corresponding to the transaction consultation message based on the state of the target object and the transaction consultation message includes: The target object's state and the transaction consultation message are input into the action determination model; The action determination model maps the state of the target object to the transaction consultation message and outputs the action corresponding to the transaction consultation message.

5. The method according to claim 1, wherein recommending the target transaction to the target object includes any one of the following: Send the interaction card for the target transaction to the terminal used by the target object; The media resources of the target transaction are sent to the terminal used by the target object, and the media resources are used to introduce the target transaction; Send a virtual object control instruction to the terminal used by the target object, the virtual object control instruction being used to instruct the terminal to control the virtual object to introduce the target transaction.

6. The method according to claim 1, before determining the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object, the method further includes: Obtain the object information of the target object and its interactive behavior information on the session interface; Based on the object information of the target object and the interaction behavior information, the object characteristics of the target object are determined.

7. The method according to claim 1, wherein the method for generating the plurality of preset transaction tags includes: The introductory texts of multiple preset transactions are structured to obtain the tags for the multiple preset transactions.

8. The method according to claim 1, before performing interest extraction on the plurality of historical transaction consultation messages to obtain multiple points of interest in the plurality of historical transaction consultation messages, the method further includes: Intent identification is performed on the multiple historical transaction consultation messages to obtain multiple intents corresponding to the multiple historical transaction consultation messages; If any of the multiple intentions is the target intention, the step of extracting interest from the multiple historical transaction consultation messages is performed to obtain multiple points of interest in the multiple historical transaction consultation messages.

9. The method according to claim 1, after determining the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object, the method further includes: If the action is to reply to the transaction inquiry message, determine the transaction tag corresponding to the transaction inquiry message; The response message to the transaction inquiry message is determined based on the transaction tag.

10. A transaction recommendation device, the device comprising: The message acquisition module is used to acquire transaction consultation messages from the target object; The action determination module is used to determine the action corresponding to the transaction consultation message based on the transaction consultation message, multiple historical transaction consultation messages of the target object, and the object characteristics of the target object. The action includes replying to the transaction consultation message and recommending a transaction. The transaction determination module is used to determine the target transaction from multiple candidate transactions based on the transaction consultation message when the action is a recommended transaction. The multiple candidate transactions are obtained by continuous recall. The specific method of continuous recall is to perform transaction recall based on each historical transaction consultation message after receiving it. The recommendation module is used to recommend the target transaction to the target object; The transaction determination module is used to extract interests from the transaction consultation message to obtain the interest points in the transaction consultation message; and to determine the target transaction from the plurality of candidate transactions based on the interest points in the transaction consultation message. The transaction determination module is used to map the points of interest in the transaction consultation message to obtain the transaction tags corresponding to the points of interest in the transaction consultation message; and to filter the multiple candidate transactions using the transaction tags to obtain the target transaction, wherein the target transaction has the transaction tags. The device further includes: The recall module is used to extract interests from the multiple historical transaction consultation messages to obtain multiple points of interest in the multiple historical transaction consultation messages; and to perform multi-way recall based on the multiple points of interest in the multiple historical transaction consultation messages to obtain the multiple candidate transactions, wherein multi-way recall refers to recalling in different ways. The recall module is used to perform tag mapping on the multiple points of interest to obtain multiple transaction tags corresponding to the multiple points of interest. The transaction tags include at least one of transaction type, transaction attribute, and transaction individual. The multiple transaction tags are matched with multiple preset transactions to obtain the multiple candidate transactions.

11. A computer device comprising one or more processors and one or more memories, wherein at least one computer program is stored in the one or more memories, the computer program being loaded and executed by the one or more processors to implement the transaction recommendation method as claimed in any one of claims 1 to 9.

12. A computer-readable storage medium storing at least one computer program, the computer program being loaded and executed by a processor to implement the transaction recommendation method as described in any one of claims 1 to 9.

13. A computer program product comprising a computer program that, when executed by a processor, implements the transaction recommendation method according to any one of claims 1 to 9.

Images