Intent prediction method, intent prediction device, electronic device and storage medium

Abstract

This application provides an intent prediction method, intent prediction device, electronic device, and storage medium, belonging to the field of artificial intelligence technology. The method includes: acquiring a sequence of object behaviors of a target object, wherein the object behavior sequence includes object behavior data and a target timestamp of the object behavior data; embedding the object behavior data to obtain target behavior type features and target behavior value features; embedding the target timestamp to obtain target time embedding features; concatenating the target behavior type features, target behavior value features, and target time embedding features to obtain a target object behavior vector; performing semantic extraction on the target object behavior vector to obtain a behavior semantic vector corresponding to the object behavior sequence; and performing intent prediction based on a preset function and the behavior semantic vector to obtain target prediction data, which is used to characterize whether the target object has a complaint intention. This application can improve the accuracy of complaint intention prediction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an intent prediction method, intent prediction device, electronic device, and storage medium. Background Technology

[0002] Currently, when predicting user intentions, such as in insurance marketing and product recommendation, most methods rely on expert experience to infer whether a user intends to complain. This approach is often highly subjective and cannot accurately predict the complaint intentions of different users. Therefore, improving the accuracy of complaint intention prediction has become an urgent technical problem to be solved. Summary of the Invention

[0003] The main objective of this application is to provide an intent prediction method, intent prediction device, electronic device, and storage medium, which aim to improve the accuracy of complaint intent prediction.

[0004] To achieve the above objectives, a first aspect of this application provides an intent prediction method, the method comprising:

[0005] Obtain the object behavior sequence of the target object, wherein the object behavior sequence includes object behavior data and the target timestamp of the object behavior data;

[0006] The object behavior data is embedded to obtain target behavior type features and target behavior value features of the target behavior type features;

[0007] The target timestamp is embedded to obtain the target time embedding feature;

[0008] The target behavior type feature, the target behavior value feature, and the target time embedding feature are concatenated to obtain the target object behavior vector;

[0009] Semantic extraction is performed on the target object's behavior vector to obtain the behavior semantic vector corresponding to the object's behavior sequence;

[0010] Intent prediction is performed based on a preset function and the behavioral semantic vector to obtain target prediction data, which is used to characterize whether the target object has a complaint intention.

[0011] In some embodiments, the object behavior data includes preservation operation data, the target behavior type feature includes a first behavior type feature, and the target behavior value feature includes a first behavior value feature of the first behavior type feature. The embedding process of the object behavior data to obtain the target behavior type feature and the target behavior value feature of the target behavior type feature includes:

[0012] The preservation operation data is embedded to obtain the first behavior type feature;

[0013] Based on a preset feature lookup table, the first behavior value feature corresponding to the first behavior type feature is determined.

[0014] In some embodiments, the object behavior data includes query content data, the target behavior type feature includes a second behavior type feature, and the target behavior value feature includes a second behavior value feature of the second behavior type feature. The embedding process of the object behavior data to obtain the target behavior type feature and the target behavior value feature of the target behavior type feature includes:

[0015] Extract the type tags corresponding to the query content data, and embed the category tags to obtain the second behavior type feature;

[0016] Based on a preset character dictionary, the text characters in the query content data are embedded to obtain the embedding representation features of each text character;

[0017] Obtain the total number of text characters in the queried content data;

[0018] Based on the embedding representation features and the total number of text characters, text semantic embedding features are obtained, and the text semantic embedding features are used as the second behavior value features.

[0019] In some embodiments, the step of semantically extracting the target object behavior vector to obtain the behavior semantic vector corresponding to the object behavior sequence includes:

[0020] Semantic encoding is performed on the behavior vector of the target object to obtain behavioral semantic encoding features;

[0021] The behavioral semantic encoding features are subjected to average pooling to obtain the behavioral semantic vector.

[0022] In some embodiments, the target prediction data includes first prediction data and second prediction data, and the step of performing intent prediction based on a preset function and the behavioral semantic vector to obtain the target prediction data includes:

[0023] Intent scoring is performed based on a preset function and the behavioral semantic vector to obtain the object's intent score;

[0024] If the object's intent score is greater than a preset intent score threshold, then the first prediction data is obtained. The first prediction data is used to indicate that the target object has a complaint intention.

[0025] If the object's intent score is less than or equal to the intent score threshold, then the second prediction data is obtained. The second prediction data is used to indicate that the target object does not have a complaint intention.

[0026] In some embodiments, after performing intent prediction based on a preset function and the behavioral semantic vector to obtain target prediction data, the method includes:

[0027] If the target prediction data is the first prediction data, then follow-up strategy information is generated based on the basic information of the target object and the object behavior sequence, wherein the first prediction data is used to indicate that the target object has a complaint intention;

[0028] The return visit strategy information is sent to the feedback recipient.

[0029] In some embodiments, obtaining the object behavior sequence of the target object includes:

[0030] Obtain historical behavior data of the target object, wherein the historical behavior data includes the historical behavior of the target object and the historical timestamp of the historical behavior;

[0031] The historical behaviors are filtered based on the historical timestamps and preset time intervals to obtain the object behavior sequence.

[0032] To achieve the above objectives, a second aspect of this application provides an intent prediction apparatus, the apparatus comprising:

[0033] The behavior sequence acquisition module is used to acquire the object behavior sequence of the target object, wherein the object behavior sequence includes object behavior data and the target timestamp of the object behavior data;

[0034] The behavior data embedding module is used to embed the object behavior data to obtain target behavior type features and target behavior value features of the target behavior type features;

[0035] The timestamp embedding module is used to embed the target timestamp to obtain the target timestamp embedding feature;

[0036] The concatenation module is used to concatenate the target behavior type feature, the target behavior value feature, and the target time embedding feature to obtain the target object behavior vector;

[0037] The semantic extraction module is used to extract semantics from the target object's behavior vector to obtain the behavior semantic vector corresponding to the object's behavior sequence.

[0038] The intent prediction module is used to predict intent based on a preset function and the behavioral semantic vector to obtain target prediction data, which is used to characterize whether the target object has a complaint intention.

[0039] To achieve the above objectives, a third aspect of the present application provides an electronic device, the electronic device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method described in the first aspect.

[0040] To achieve the above objectives, a fourth aspect of the present application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in the first aspect.

[0041] The intent prediction method, device, electronic device, and storage medium proposed in this application obtain an object behavior sequence of a target object, wherein the object behavior sequence includes object behavior data and target timestamps of the object behavior data, thereby obtaining an object behavior sequence for intent prediction, and thus realizing the prediction of whether the target object has a complaint intention based on the target object's operational behavior. Further, the object behavior data is embedded to obtain target behavior type features and target behavior value features, and the target timestamp is embedded to obtain target time embedding features; then, the target behavior type features, target behavior value features, and target time embedding features are concatenated to obtain a target object behavior vector. This allows for the construction of target object behavior vectors based on different object behaviors and the timestamps of each object behavior, making the feature content information of the target object behavior vectors more complete. Furthermore, semantic extraction is performed on the target object behavior vectors to obtain behavioral semantic vectors corresponding to the object behavior sequence, which can conveniently extract the behavioral semantic content from the object behavior data. Finally, based on the preset function and behavioral semantic vector, intent prediction is performed to obtain target prediction data. It can clearly determine whether the target object has a complaint intention based on the target object's behavioral semantic vector. It can predict the magnitude of the complaint intention based on the target object's behavior, and can significantly improve the accuracy of predicting the target object's complaint intention. Attached Figure Description

[0042] Figure 1 This is a flowchart of the intent prediction method provided in the embodiments of this application;

[0043] Figure 2 yes Figure 1 The flowchart of step S101 in the text;

[0044] Figure 3 yes Figure 1 The flowchart of step S102 in the document;

[0045] Figure 4 yes Figure 1 Another flowchart of step S102 in the process;

[0046] Figure 5 yes Figure 1 The flowchart of step S105 in the process;

[0047] Figure 6 yes Figure 1 The flowchart of step S106 in the process;

[0048] Figure 7 This is another flowchart of the intent prediction method provided in the embodiments of this application;

[0049] Figure 8 This is a schematic diagram of the intent prediction device provided in the embodiments of this application;

[0050] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0052] It should be noted that although functional modules are divided in the device schematic diagram and a logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first," "second," etc., in the specification, claims, and the aforementioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit this application.

[0054] First, let's analyze some of the terms used in this application:

[0055] Artificial intelligence (AI) is a new branch of computer science that studies, develops, and applies theories, methods, technologies, and systems to simulate, extend, and expand human intelligence. It aims to understand the essence of intelligence and produce intelligent machines that can react in a way similar to human intelligence. Research in this field includes robotics, speech recognition, image recognition, natural language processing, and expert systems. AI can simulate the information processes of human consciousness and thought. Furthermore, AI utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to achieve optimal results.

[0056] Natural Language Processing (NLP): NLP uses computers to process, understand, and utilize human language (such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary field of computer science and linguistics, often referred to as computational linguistics. NLP includes syntactic analysis, semantic analysis, and discourse understanding. It is commonly used in machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, intent recognition, information extraction and filtering, text classification and clustering, sentiment analysis, and opinion mining. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computation.

[0057] Information Extraction (NER) is a text processing technique that extracts factual information such as entities, relationships, and events from natural language text and outputs it as structured data. Information extraction is a technique for extracting specific information from text data. Text data is composed of specific units, such as sentences, paragraphs, and chapters; text information is also composed of specific units, such as characters, words, phrases, sentences, paragraphs, or combinations of these units. Extracting noun phrases, names of people, and place names from text data is an example of text information extraction. Of course, text information extraction techniques can extract information of various types.

[0058] A timestamp is a sequence of characters or encoded information used to mark when a specific event occurred, usually given as a date and time.

[0059] Currently, when predicting user intentions, such as in insurance marketing and product recommendation, most methods rely on expert experience to infer whether a user intends to complain. This approach is often highly subjective and cannot accurately predict the complaint intentions of different users. Therefore, improving the accuracy of complaint intention prediction has become an urgent technical problem to be solved.

[0060] Based on this, embodiments of this application provide an intent prediction method, intent prediction device, electronic device, and storage medium, aiming to improve the accuracy of complaint intent prediction.

[0061] The intent prediction method, intent prediction device, electronic device, and storage medium provided in the embodiments of this application are specifically described through the following embodiments. First, the intent prediction method in the embodiments of this application is described.

[0062] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, 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 obtain optimal results.

[0063] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0064] The intent prediction method provided in this application relates to the field of artificial intelligence technology. The intent prediction method provided in this application can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent 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, CDN, and big data and artificial intelligence platforms; the software can be an application implementing the intent prediction method, but is not limited to the above forms.

[0065] This application can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0066] It should be noted that in all specific embodiments of this application, when processing data related to user identity or characteristics, such as user information, user behavior data, user historical data, and user location information, user permission or consent is obtained first. Furthermore, the collection, use, and processing of this data comply with relevant laws, regulations, and standards of the relevant countries and regions. In addition, when embodiments of this application require access to sensitive personal information of users, separate permission or consent from the user is obtained through pop-ups or redirects to confirmation pages. Only after obtaining the user's separate permission or consent is the necessary user-related data for the proper functioning of the embodiments of this application obtained.

[0067] Figure 1 This is an optional flowchart of the intent prediction method provided in the embodiments of this application. Figure 1 The method may include, but is not limited to, steps S101 to S106.

[0068] Step S101: Obtain the object behavior sequence of the target object, wherein the object behavior sequence includes object behavior data and the target timestamp of the object behavior data;

[0069] Step S102: Embed the object behavior data to obtain the target behavior type feature and the target behavior value feature of the target behavior type feature;

[0070] Step S103: Embed the target timestamp to obtain the target time embedding feature;

[0071] Step S104: Concatenate the target behavior type feature, target behavior value feature, and target time embedding feature to obtain the target object behavior vector;

[0072] Step S105: Semantic extraction is performed on the target object behavior vector to obtain the behavior semantic vector corresponding to the object behavior sequence;

[0073] Step S106: Based on the preset function and behavioral semantic vector, intention prediction is performed to obtain target prediction data. The target prediction data is used to characterize whether the target object has a complaint intention.

[0074] Steps S101 to S106 of this embodiment of the application involve obtaining a sequence of object behavior of a target object. This sequence includes object behavior data and target timestamps of the object behavior data. This allows for the prediction of whether the target object has a complaint intention based on its operational behavior. Further, the object behavior data is embedded to obtain target behavior type features and target behavior value features. The target timestamps are also embedded to obtain target time embedding features. These features are then concatenated to obtain a target object behavior vector. This allows for the construction of target object behavior vectors based on different object behaviors and the timestamps of each behavior, resulting in more complete feature content information for the target object behavior vectors. Furthermore, semantic extraction is performed on the target object behavior vectors to obtain behavioral semantic vectors corresponding to the object behavior sequence. This facilitates the extraction of behavioral semantic content from the object behavior data. Finally, based on the preset function and behavioral semantic vector, intent prediction is performed to obtain target prediction data. It can clearly determine whether the target object has a complaint intention based on the target object's behavioral semantic vector. It can predict the magnitude of the complaint intention based on the target object's behavior, and can significantly improve the accuracy of predicting the target object's complaint intention.

[0075] Please see Figure 2 In some embodiments, step S101 may include, but is not limited to, steps S201 to S202:

[0076] Step S201: Obtain historical behavior data of the target object, wherein the historical behavior data includes the historical behavior of the target object and the historical timestamp of the historical behavior;

[0077] Step S202: Filter historical behaviors based on historical timestamps and preset time intervals to obtain object behavior sequences.

[0078] In step S201 of some embodiments, a web crawler can be written, and after setting up the data source, data can be crawled in a targeted manner to obtain the historical behavior data of the target object. Other methods can also be used to obtain the historical behavior data of the target object, and are not limited to these. The target object can be an insurance user, etc., and the historical behavior data includes the target object's historical behaviors and the corresponding historical timestamps. Further, the historical behaviors in the historical behavior data include the target object's policy maintenance operations and search query behaviors, and the historical timestamp is the time value corresponding to the target object performing a policy maintenance operation or a search query behavior. "Policy maintenance" is a general term in the insurance field, and policy maintenance operations include adding or reducing coverage, adding supplementary insurance, etc. The historical behavior data corresponding to policy maintenance operations is a discrete feature data. The historical behavior data corresponding to search query behaviors mainly consists of insurance-related content searched by the target object on mini-programs, online platforms, or other devices; this historical behavior data corresponding to search query behaviors is text feature data.

[0079] In step S202 of some embodiments, in order to improve the effectiveness and rationality of intent prediction, historical behaviors can be filtered according to historical timestamps and preset time intervals. For example, the preset time interval can be set to the past week or the past month, etc. Based on the historical timestamps of historical behaviors, historical behaviors with historical timestamps within the preset time interval are selected as valid behaviors. The valid behaviors are used as object behavior data and are denoted as B. Each object behavior data B corresponds to a specific operation data C (including preservation operation data or query content data) and a target timestamp T. All valid behaviors are sorted according to the order of timestamps to generate an object behavior sequence of the target object. The object behavior sequence includes object behavior data and the target timestamp of the object behavior data. The object behavior sequence can be expressed as the following formula (1):

[0080] B = [B0, B1, ..., B] L-1 ]=[[T0,C0],[T1,C1],……,[T L-1 C L-1 Equation (1)

[0081] Among them, B i =[T i C i ], B i For the behavior data (i.e., valid behavior) of the i-th object, T i C represents the occurrence time (i.e., the target timestamp) of the i-th object's behavior data. i This refers to the specific operation data corresponding to the i-th object's behavior data. The object behavior sequence is ordered chronologically, B. L-1It is the last object behavior data that occurred, and L is the length of the object behavior sequence.

[0082] Through the above steps S201 to S202, the object behavior data of the target object and the timestamp corresponding to each object behavior data can be determined relatively easily, and the object behavior sequence used for intention prediction can be obtained, thereby realizing the prediction of whether the target object has a complaint intention by using the operation behavior of the target object.

[0083] Please see Figure 3 In some embodiments, the object behavior data includes preservation operation data, the target behavior type feature includes a first behavior type feature, the target behavior value feature includes a first behavior value feature of the first behavior type feature, and step S102 may include, but is not limited to, steps S301 to S302:

[0084] Step S301: Embed the data of the preservation operation to obtain the first behavior type feature;

[0085] Step S302: Based on the preset feature lookup table, determine the first row value feature corresponding to the first row type feature.

[0086] In step S301 of some embodiments, object behavior data can be input into a preset intent prediction model. This intent prediction model can be constructed based on a convolutional neural network model and includes an embedding layer, an encoding layer, and an output layer. The embedding layer is mainly used to embed the input behavior data to obtain behavior type features and behavior value features corresponding to the behavior data. The encoding layer is used to generate semantic representation information corresponding to the behavior data based on the obtained behavior type features and behavior value features. Finally, the output layer performs intent prediction on the behavior data based on the semantic representation information, thereby determining whether the behavior indicates a complaint intent. Therefore, object behavior data containing preservation operation data is input into the preset intent prediction model. Based on the embedding layer, the preservation operation data is mapped from the data space to the vector space to obtain the first behavior type feature corresponding to the preservation operation data. This first behavior type feature contains feature information corresponding to the preservation operation.

[0087] In step S302 of some embodiments, since the insurance maintenance operation data includes various specific operations such as increasing insurance coverage, decreasing insurance coverage, adding supplementary insurance, modifying object information, and supplementing disclosure instructions, and different specific operations are preset with corresponding behavior values, a feature lookup table can be pre-constructed based on the specific operations and behavior values. Through the feature lookup table, the behavior value of the specific operation represented by the first behavior type feature after mapping processing is queried, and this behavior value is represented in vector form to obtain the first behavior value feature of the first behavior type feature.

[0088] For example, if the specific operation of the current policy maintenance data is to add supplementary insurance, then this specific operation is mapped from the data space to the vector space. The resulting vector feature is the first behavior type feature corresponding to this specific operation. Then, by querying the feature lookup table, the behavior value corresponding to the specific operation of adding supplementary insurance is found. This behavior value is converted from numerical form to vector form, which is the first behavior value feature corresponding to the first behavior type feature.

[0089] Through the above steps S301 to S302, the security operation behavior of the target object can be embedded relatively easily, and the key behavioral feature information in the security operation behavior can be extracted. This allows the behavioral feature information of the security operation to be used for intention prediction in subsequent intention prediction, thereby improving the prediction accuracy.

[0090] Please see Figure 4 In some embodiments, the object behavior data includes query content data, the target behavior type feature includes a second behavior type feature, the target behavior value feature includes a second behavior value feature of the second behavior type feature, and step S102 may also include, but is not limited to, steps S401 to S404:

[0091] Step S401: Extract the type labels corresponding to the query content data, and embed the category labels to obtain the second row type features;

[0092] Step S402: Based on a preset character dictionary, perform an embedding representation query on the text characters in the query content data to obtain the embedding representation features of each text character;

[0093] Step S403: Obtain the total number of text characters in the query content data;

[0094] Step S404: Based on the embedding representation features and the total number of text characters, obtain the text semantic embedding features, and use the text semantic embedding features as the second row value features.

[0095] In step S401 of some embodiments, the type labels of the query content data include insurance type, policy amount range, etc. The object behavior data containing the query content data is input into the aforementioned intent prediction model. Keyword recognition is performed on the query content data based on the embedding layer to obtain query keywords. Multiple preset candidate type labels are then filtered based on the query keywords to determine the type label corresponding to the query content data. This type label is then embedded, mapping it from the data space to the vector space to obtain the second behavior type feature corresponding to that type label.

[0096] In step S402 of some embodiments, the query content data is segmented into words, splitting the query content data from text form into character form to obtain multiple text characters. Then, a preset character dictionary is used to perform an embedding representation query on each text character to determine the embedding representation vector of each text character, thereby obtaining the embedding representation feature of each text character. This embedding representation feature can characterize the semantic information of each text character. The preset character dictionary V... text It can be constructed based on expert experience, etc., without any restrictions. This character dictionary includes multiple pairs of text characters and their corresponding embedding vectors. The embedding vector of each text character can be initialized using a pre-defined text model (e.g., BERT model, word2vec model). The pre-defined character dictionary V... text The size is (|V text |,H), where H is the length of the vector representing the embedded feature representation.

[0097] In step S403 of some embodiments, statistical functions or other statistical tools are used to count the total number of text characters in the query content data to determine the total number of text characters contained in the query content data, thus obtaining the total number of text characters.

[0098] In step S404 of some embodiments, the embedding representation features of all text characters in the query content data are first summed to obtain the text content embedding feature. Then, the text content embedding feature is divided by the total number of text characters to obtain the text semantic embedding feature. This text semantic embedding feature is then used as the second row value feature e corresponding to the second row type feature. value .

[0099] It should be noted that the process of obtaining text semantic embedding features based on embedding representation features and the total number of text characters can be expressed as follows (2):

[0100]

[0101] Among them, e query Text semantic embedding features (i.e., second-row value features e) value ), where query is the query content data of the target object. j E represents the j-th text character of the query content data of the target object. text len is a predefined character dictionary. query For the total number of text characters, lookup(k,V) text ) represents the embedding feature representation of text character k.

[0102] Through the above steps S401 to S404, the query operation behavior of the target object can be embedded relatively easily, and the more critical query content information in the query operation behavior can be extracted. This allows the query content information of the query operation to be used for intent prediction in subsequent intent prediction, thereby improving the prediction accuracy.

[0103] In step S103 of some embodiments, the target timestamp is used to represent the time when the object behavior data occurs. The target timestamp is input into the aforementioned intent prediction model, and when the target timestamp is embedded in the embedding layer to obtain the target time embedding feature, the time difference between the current object behavior data and the last object behavior data can be modeled, and this time difference is used as the embedded representation of the target timestamp. Specifically, the process of embedding the target timestamp to obtain the target time embedding feature can be expressed as follows (3):

[0104]

[0105] in, T is the target temporal embedding feature of the i-th object behavior data. L-1 -T i E represents the time difference between the time of the i-th object's behavior data and the time of the last object's behavior data. time The preset time vector can be randomly generated. Since the preset time interval used in this embodiment is one month, the maximum time difference is one month. If the preset time interval is another time period, it can be modified accordingly. This method makes it relatively easy to determine the target time embedding feature of the target timestamp corresponding to each object's behavioral data. This allows the intent prediction model to learn the timestamp information of each object's behavioral data when predicting the complaint intention of the target object, automatically learning and modeling the chronological order of each object's behavioral data, thereby improving the accuracy of intent prediction.

[0106] In step S104 of some embodiments, when concatenating the target behavior type feature, target behavior value feature, and target time embedding feature, the target behavior type feature, target behavior value feature, and target time embedding feature corresponding to each object behavior data can be directly vector-added to obtain the target object behavior vector corresponding to each object behavior data. For example, the target object behavior vector can be represented as:

[0107] E = [e0, e1, ..., e i ]

[0108] in, e i Let i be the target object behavior vector of the i-th object behavior data. For the target temporal embedding feature of the i-th object's behavior data, For the target behavior type feature of the i-th object behavior data, Let i be the target behavior value feature of the i-th object's behavior data.

[0109] Please see Figure 5 In some embodiments, step S105 may include, but is not limited to, steps S501 to S502:

[0110] Step S501: Semantically encode the target object's behavior vector to obtain behavior semantic encoding features;

[0111] Step S502: Perform average pooling on the behavioral semantic encoding features to obtain the behavioral semantic vector.

[0112] In step S501 of some embodiments, the target object behavior vector can be input into the encoding layer of the aforementioned intent prediction model. Semantic extraction is then performed on the target object behavior vector based on the encoding layer to obtain the behavior semantic vector corresponding to the object behavior sequence. Specifically, the encoding layer is used to perform semantic extraction on the target object behavior vector using a transformer structure to obtain semantic feature information from the target object behavior vector, resulting in behavior semantic encoding feature A. For example, the process of semantic extraction of the target object behavior vector E using the encoding layer can be represented as A = transformer(E), and the behavior semantic encoding feature can be represented as A = [A0, A1, ..., A...]. i ,…A L-1 ], where A i The behavioral semantic encoding feature is the behavioral data corresponding to the i-th object.

[0113] In step S502 of some embodiments, the behavioral semantic coding features are averaged using the coding layer to extract the comprehensive semantic information of all semantic coding features, resulting in a behavioral semantic vector S. This process can be represented by the following equation (4):

[0114]

[0115] Among them, S i It is the behavior semantic vector corresponding to the i-th object behavior data. Based on this method, the behavior semantic vector of each object behavior data is calculated. The average of all the behavior semantic vectors is then calculated to obtain the final behavior semantic vector S of the target object.

[0116] Through the above steps S501 to S502, the behavioral semantic content in the object behavior data can be extracted relatively easily to obtain the behavioral semantic vector of the target object, which enables the use of the behavioral semantic vector of the target object to perform intent prediction and improve the accuracy of intent prediction.

[0117] Please see Figure 6 In some embodiments, the target prediction data includes first prediction data and second prediction data, and step S106 includes, but is not limited to, steps S601 to S603:

[0118] Step S601: Based on the preset function and the behavioral semantic vector, perform intent scoring to obtain the object intent score;

[0119] Step S602: If the object's intent score is greater than a preset intent score threshold, then the first prediction data is obtained. The first prediction data is used to indicate that the target object has a complaint intention.

[0120] Step S603: If the object's intent score is less than or equal to the intent score threshold, then the second prediction data is obtained. The second prediction data is used to indicate that the target object does not have a complaint intention.

[0121] In step S601 of some embodiments, the behavioral semantic vector is input into the output layer of the above-mentioned intent prediction model. The preset function in the output layer can be a softmax function or a sigmoid function, etc., without limitation. Taking the sigmoid function as an example, the intent score is obtained by using the sigmoid function and the behavioral semantic vector. The intent scoring process can be expressed as follows (5):

[0122] p = sigmoid(<S,W> +b) Equation (5)

[0123] Where p is the object intent score, S is the behavioral semantic vector of the target object, and W and b are both learnable model parameters, where W is a vector.<S,W> To perform a dot product between the behavioral semantic vector S and the parameter W.

[0124] In step S602 of some embodiments, since the object intent score can clearly reflect the strength of the target object's complaint intent—that is, the higher the object intent score, the greater the target object's complaint intent, and the lower the object intent score, the smaller the target object's complaint intent—it can be determined whether the target object has a complaint intent based on the object intent score. Specifically, the object intent score is compared with a preset intent score threshold, where the specific value of the intent score threshold can be set according to actual conditions and is not limited; for example, the intent score threshold can be 0.5. If the object intent score is greater than the preset intent score threshold, it indicates that the target object has a relatively large complaint intent, thus obtaining first predicted data, which indicates that the target object has a complaint intent.

[0125] In step S603 of some embodiments, if the object intent score is less than or equal to a preset intent score threshold, it indicates that the target object has a low intention to complain, and thus a second prediction data is obtained, which indicates that the target object does not have an intention to complain.

[0126] Through the above steps S601 to S603, it is possible to clearly determine whether the target object has a complaint intention based on the target object's behavioral semantic vector. It is possible to predict the magnitude of the complaint intention based on the target object's object behavior, and quantify the probability that the target object has a complaint intention by using the object intention score. This can significantly improve the accuracy of predicting the target object's complaint intention.

[0127] Please see Figure 7 After step S106 in some embodiments, the intent prediction method may also include, but is not limited to, steps S701 to S702:

[0128] Step S701: If the target prediction data is the first prediction data, then generate follow-up strategy information based on the basic information of the target object and the object behavior sequence, wherein the first prediction data is used to indicate that the target object has the intention to complain;

[0129] Step S702: Send the follow-up strategy information to the feedback recipient.

[0130] In step S701 of some embodiments, if the target prediction data is the first prediction data, it indicates that the target object has the intention to complain. At this time, it is necessary to conduct a follow-up survey on the target object in a timely manner. Therefore, the basic information and object behavior sequence of the target object can be extracted from a preset database. The basic information includes the target object's age, gender, occupation, etc., and the object behavior sequence is the object behavior data obtained in step S101 above. Based on the basic information and object behavior sequence, multiple preset strategy information is filtered, and the strategy information that best matches the target object is selected as the follow-up strategy information. The follow-up strategy information includes the dialogue content, dialogue script, etc. used for the follow-up.

[0131] In step S702 of some embodiments, the follow-up strategy information is sent to the feedback recipient through a preset network platform or email and other means. The feedback recipient includes insurance staff or other relevant personnel.

[0132] Through the above steps S701 to S702, the feedback recipient can conduct a follow-up visit to the target recipient based on the follow-up strategy information to understand the target recipient's current needs and the target recipient's true intention for the current service. In this way, corresponding problem-solving measures can be taken to address the target recipient's needs, improve the current business service, and enhance the target recipient's experience.

[0133] The intent prediction method of this application embodiment obtains a sequence of object behavior of a target object, wherein the object behavior sequence includes object behavior data and target timestamps of the object behavior data, thereby obtaining an object behavior sequence for intent prediction, and thus realizing the prediction of whether the target object has a complaint intention by utilizing the target object's operational behavior. Further, the object behavior data is embedded to obtain target behavior type features and target behavior value features, and the target timestamp is embedded to obtain target time embedding features; then, the target behavior type features, target behavior value features, and target time embedding features are concatenated to obtain a target object behavior vector. This allows for the construction of target object behavior vectors based on different object behaviors and the timestamps of each object behavior, making the feature content information of the target object behavior vectors more complete. Further, semantic extraction is performed on the target object behavior vectors to obtain behavioral semantic vectors corresponding to the object behavior sequence, which can conveniently extract the behavioral semantic content from the object behavior data. Finally, based on the preset function and behavioral semantic vector, intent prediction is performed to obtain target prediction data. It can clearly determine whether the target object has a complaint intention based on the target object's behavioral semantic vector. It can predict the magnitude of the complaint intention based on the target object's behavior, and can significantly improve the accuracy of predicting the target object's complaint intention.

[0134] Please see Figure 8 This application also provides an intent prediction apparatus that can implement the above-described intent prediction method. The apparatus includes:

[0135] The behavior sequence acquisition module 801 is used to acquire the object behavior sequence of the target object, wherein the object behavior sequence includes object behavior data and the target timestamp of the object behavior data;

[0136] The behavior data embedding module 802 is used to embed object behavior data to obtain target behavior type features and target behavior value features of target behavior type features;

[0137] The timestamp embedding module 803 is used to embed the target timestamp to obtain the target time embedding feature;

[0138] The splicing module 804 is used to splice the target behavior type features, target behavior value features, and target time embedding features to obtain the target object behavior vector;

[0139] The semantic extraction module 805 is used to extract semantics from the target object's behavior vector to obtain the behavior semantic vector corresponding to the object's behavior sequence;

[0140] The intent prediction module 806 is used to predict intent based on a preset function and behavioral semantic vector to obtain target prediction data. The target prediction data is used to characterize whether the target object has a complaint intention.

[0141] The specific implementation of this intent prediction device is basically the same as the specific embodiment of the intent prediction method described above, and will not be repeated here.

[0142] This application also provides an electronic device, which includes: a memory, a processor, a program stored in the memory and executable on the processor, and a data bus for communication between the processor and the memory. When the program is executed by the processor, it implements the aforementioned intent prediction method. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0143] Please see Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is illustrated. The electronic device includes:

[0144] The processor 901 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this application.

[0145] The memory 902 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store the operating system and other applications. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called and executed by the processor 901 using the intent prediction method of the embodiments of this application.

[0146] The input / output interface 903 is used to implement information input and output;

[0147] The communication interface 904 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0148] Bus 905 transmits information between various components of the device (e.g., processor 901, memory 902, input / output interface 903, and communication interface 904);

[0149] The processor 901, memory 902, input / output interface 903, and communication interface 904 are connected to each other within the device via bus 905.

[0150] This application also provides a computer-readable storage medium storing one or more programs that can be executed by one or more processors to implement the above-described intent prediction method.

[0151] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0152] The intent prediction method, intent prediction device, electronic device, and computer-readable storage medium provided in this application obtain an object behavior sequence of a target object, wherein the object behavior sequence includes object behavior data and target timestamps of the object behavior data. This allows for the prediction of whether the target object has a complaint intention based on its operational behavior. Further, the object behavior data is embedded to obtain target behavior type features and target behavior value features, and the target timestamp is embedded to obtain target time embedding features. The target behavior type features, target behavior value features, and target time embedding features are then concatenated to obtain a target object behavior vector. This allows for the construction of target object behavior vectors based on different object behaviors and the timestamps of each object behavior, resulting in more complete feature content information in the target object behavior vectors. Furthermore, semantic extraction is performed on the target object behavior vectors to obtain behavioral semantic vectors corresponding to the object behavior sequence, which facilitates the extraction of behavioral semantic content from the object behavior data. Finally, based on the preset function and behavioral semantic vector, intent prediction is performed to obtain target prediction data. It can clearly determine whether the target object has a complaint intention based on the target object's behavioral semantic vector. It can predict the magnitude of the complaint intention based on the target object's behavior, and can significantly improve the accuracy of predicting the target object's complaint intention.

[0153] The embodiments described in this application are for the purpose of more clearly illustrating the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions provided by the embodiments of this application. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this application are also applicable to similar technical problems.

[0154] It will be understood by those skilled in the art that Figure 1-7 The technical solutions shown do not constitute a limitation on the embodiments of this application, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0155] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0156] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0157] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0158] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of the units described above is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0160] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0162] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing programs, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The preferred embodiments of the present application have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present application. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and substance of the embodiments of the present application shall be within the scope of the claims of the present application.

Claims

1. An intention prediction method, characterized by, The method comprises: acquiring an object behavior sequence of a target object, wherein the object behavior sequence comprises object behavior data and a target timestamp of the object behavior data; performing embedding processing on the object behavior data to obtain a target behavior type feature and a target behavior value feature of the target behavior type feature; performing embedding processing on the target timestamp to obtain a target time embedding feature; concatenating the target behavior type feature, the target behavior value feature and the target time embedding feature to obtain a target object behavior vector; performing semantic extraction on the target object behavior vector to obtain a behavior semantic vector corresponding to the object behavior sequence; performing intention prediction based on a preset function and the behavior semantic vector to obtain target prediction data, wherein the target prediction data is used to represent whether the target object has a complaint intention; the object behavior data comprises security operation data and query content data, the target behavior type feature comprises a first behavior type feature and a second behavior type feature, and the target behavior value feature comprises a first behavior value feature of the first behavior type feature and a second behavior value feature of the second behavior type feature; the embedding processing on the object behavior data to obtain the target behavior type feature and the target behavior value feature of the target behavior type feature comprises: performing embedding processing on the security operation data to obtain the first behavior type feature; determining the first behavior value feature corresponding to the first behavior type feature based on a preset feature comparison table; extracting a type label corresponding to the query content data and performing embedding processing on the type label to obtain the second behavior type feature; embedding representing text characters in the query content data based on a preset character dictionary to obtain an embedding representation feature of each text character; obtaining a total amount of text characters of the query content data; and obtaining a text semantic embedding feature based on the embedding representation feature and the total amount of text characters, and taking the text semantic embedding feature as the second behavior value feature.

2. The intention prediction method of claim 1, wherein the semantic extraction on the target object behavior vector to obtain the behavior semantic vector corresponding to the object behavior sequence comprises: performing semantic coding on the target object behavior vector to obtain a behavior semantic coding feature; performing average pooling processing on the behavior semantic coding feature to obtain the behavior semantic vector.

3. The intent prediction method of claim 1, wherein, the target prediction data comprises first prediction data and second prediction data, and the intention prediction based on the preset function and the behavior semantic vector to obtain the target prediction data comprises: performing intention scoring based on the preset function and the behavior semantic vector to obtain an object intention score; if the object intention score is greater than a preset intention score threshold, the first prediction data is obtained, and the first prediction data is used to represent that the target object has a complaint intention; if the object intention score is less than or equal to the intention score threshold, the second prediction data is obtained, and the second prediction data is used to represent that the target object does not have a complaint intention.

4. The intention prediction method of claim 1, wherein, After the intention prediction based on the preset function and the behavior semantic vector is performed, target prediction data is obtained, the method comprises: If the target prediction data is first prediction data, generating revisit strategy information according to basic information and object behavior sequence of the target object, wherein the first prediction data is used to indicate that the target object has complaint intention; Sending the revisit strategy information to a feedback object.

5. The intention prediction method according to any one of claims 1 to 4, characterized in that, The object behavior sequence of the target object is obtained, comprising: Obtaining historical behavior data of the target object, wherein the historical behavior data comprises historical behavior of the target object and a historical timestamp of the historical behavior; Filtering the historical behavior according to the historical timestamp and a preset time interval to obtain the object behavior sequence.

6. An intention prediction apparatus characterized by comprising: The device comprises: A behavior sequence acquisition module, configured to obtain an object behavior sequence of a target object, wherein the object behavior sequence comprises object behavior data and a target timestamp of the object behavior data; A behavior data embedding module, configured to perform embedding processing on the object behavior data to obtain target behavior type features and target behavior value features of the target behavior type features; A timestamp embedding module, configured to perform embedding processing on the target timestamp to obtain target time embedding features; A splicing module, configured to splice the target behavior type features, the target behavior value features and the target time embedding features to obtain a target object behavior vector; A semantic extraction module, configured to perform semantic extraction on the target object behavior vector to obtain a behavior semantic vector corresponding to the object behavior sequence; An intention prediction module, configured to perform intention prediction based on a preset function and the behavior semantic vector to obtain target prediction data, wherein the target prediction data is used to indicate whether the target object has complaint intention; The object behavior data comprises security operation data and query content data, the target behavior type features comprise first behavior type features and second behavior type features, and the target behavior value features comprise first behavior value features of the first behavior type features and second behavior value features of the second behavior type features; The embedding processing on the object behavior data to obtain the target behavior type features and the target behavior value features of the target behavior type features comprises: Performing embedding processing on the security operation data to obtain the first behavior type features; determining the first behavior value features corresponding to the first behavior type features based on a preset feature comparison table; Extracting type labels corresponding to the query content data and performing embedding processing on the type labels to obtain the second behavior type features; embedding representing query of text characters in the query content data based on a preset character dictionary to obtain embedding representation features of each text character; obtaining a total amount of text characters of the query content data; obtaining text semantic embedding features based on the embedding representation features and the total amount of text characters, and taking the text semantic embedding features as the second behavior value features.

7. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the intention prediction method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 7. The computer program is executed by the processor to implement the intention prediction method according to any one of claims 1 to 5.

Images