User intention recognition method, device and electronic equipment

Abstract

The present disclosure provides a user intention recognition method, device and electronic equipment, and relates to the technical field of machine learning. The user intention recognition method comprises: constructing user portrait information through user behavior data of a user terminal; collecting scene behavior data of the user terminal in a target business scenario, encoding variables contained in the scene behavior data according to a feature encoding mode corresponding to the variables to determine a feature vector corresponding to the scene behavior data; training the user portrait information and the feature vector by using a deep learning algorithm to obtain a trained user intention recognition model, and predicting the intention of a to-be-recognized user terminal to the target business scenario by using the user intention recognition model. The technical solution provided by the present disclosure can recognize the user intention in the target business scenario by using the deep learning method, and can improve the recognition accuracy of the user intention.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of machine learning, in particular, to a user intention recognition method, a user intention recognition device, and an electronic device implementing the above user intention recognition method. BACKGROUND

[0002] With the development of information networks, electronic advertisements gradually become the most important form of advertisement delivery. Advertisements delivered on the Internet have more advantages than traditional advertisements in brand promotion, product sales feedback, market reaction analysis, and other aspects. Combined with artificial intelligence technology, user data can be analyzed to achieve precise advertisement delivery and maximize advertisement effectiveness.

[0003] The second type of e-commerce refers to an independent product website outside the e-commerce platform, which can be promoted through any channel that can attract traffic to generate orders and is not subject to the rules of the e-commerce platform. In the related technology of second-type e-commerce advertisement recommendation, the division rules of potential people are usually set based on business experience and the characteristics of the goods delivered. For example, for high-end skin care product advertisements, high-end mobile devices and female groups are selected for delivery. The manual experience determines the data rules, which not only limits the number of rules used, but also cannot determine the optimal parameters of each rule, resulting in poor delivery effect of second-type e-commerce advertisements.

[0004] It should be noted that the information disclosed in the above background section is only used to strengthen the understanding of the background of the present disclosure, and therefore can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present disclosure is to provide a user intention recognition method, a user intention recognition device, an electronic device, and a computer readable storage medium, thereby avoiding the identification of advertisement intention users through artificial rules to some extent, reducing the artificial time cost, and improving the accuracy of intention recognition.

[0006] According to a first aspect of the present disclosure, a user intention recognition method is provided, which comprises: constructing user portrait information through user behavior data of a user terminal; collecting scenario behavior data corresponding to a target business scenario, and encoding variables contained in the scenario behavior data according to variable type corresponding feature encoding mode to determine a feature vector corresponding to the scenario behavior data; training the user portrait information and the feature vector using a deep learning algorithm to obtain a trained user intention recognition model, so as to predict the intention of a to-be-identified user terminal to the target business scenario through the user intention recognition model.

[0007] In an example embodiment of the present disclosure, based on the foregoing embodiment, constructing user portrait information based on user behavior data of a user terminal includes: counting user behavior data of the user terminal in different time windows to obtain online time characteristics and basic attribute characteristics of the user terminal as the user portrait information.

[0008] In an example embodiment of the present disclosure, based on the foregoing embodiment, encoding the variables according to the feature encoding mode corresponding to the variables to determine the feature vector corresponding to the scenario behavior data includes: counting network connection variables included in the scenario behavior data to obtain a first vector corresponding to the network connection variables; encoding category variables included in the scenario behavior data according to a one-hot encoding mode to obtain a second vector corresponding to the category variables; and splicing the first vector and the second vector to obtain the feature vector corresponding to the scenario behavior data.

[0009] In an example embodiment of the present disclosure, based on the foregoing embodiment, encoding the category variables included in the scenario behavior data according to the one-hot encoding mode to obtain the second vector corresponding to the category variables includes: encoding category variables with two values according to the one-hot encoding mode to obtain the corresponding second vector; and performing feature induction processing on category variables with more than three values, and one-hot encoding the processed values to obtain the corresponding second vector.

[0010] In an example embodiment of the present disclosure, based on the foregoing embodiment, the method further includes: encoding missing values in the scenario behavior data according to the values of each variable included in the scenario behavior data to determine the feature vector.

[0011] In an example embodiment of the present disclosure, based on the foregoing embodiment, training the user portrait information and the feature vector using a deep learning algorithm to obtain a trained user willingness recognition model includes: splicing the user portrait information and the feature vector to obtain a training sample; determining a willingness label corresponding to the training sample according to the scenario behavior data; and training a deep learning model using the training sample and the corresponding willingness label to obtain the trained user willingness recognition model.

[0012] In an example embodiment of the present disclosure, based on the foregoing embodiment, the target business scenario includes a second-class e-commerce advertisement, and the predicting, by the user willingness recognition model, a willingness of a user terminal to be identified to the target business scenario includes: inputting user behavior data of the user terminal to be identified into the user willingness recognition model to predict, by the user willingness recognition model, a probability of the user terminal to be identified clicking on the second-class e-commerce advertisement.

[0013] In an example embodiment of the present disclosure, based on the foregoing embodiment, the method further comprises: when the target business scenario changes, updating the scenario behavior data corresponding to the changed scenario; and reconstructing a will recognition model using the user portrait information and the updated scenario behavior data to identify the intent of the to-be-identified user terminal to the changed scenario.

[0014] According to a second aspect of the present disclosure, a user will recognition device is provided, which comprises:

[0015] a user portrait module configured to construct user portrait information based on user behavior data of a user terminal; a feature representation module configured to collect scenario behavior data corresponding to a target business scenario, encode variables included in the scenario behavior data according to a variable corresponding feature encoding mode, and determine a feature vector corresponding to the scenario behavior data; and a model application module configured to train the user portrait information and the feature vector using a deep learning algorithm, obtain a trained user will recognition model, and predict the will of a to-be-identified user terminal to the target business scenario through the user will recognition model.

[0016] In an example embodiment of the present disclosure, based on the foregoing embodiment, the user portrait module can be configured to: count user behavior data of the user terminal in different time windows to obtain online time features and basic attribute features of the user terminal as the user portrait information.

[0017] In an example embodiment of the present disclosure, based on the foregoing embodiment, the feature representation module can comprise a first encoding module, a second encoding module, and a feature splicing module.

[0018] The first encoding module can be configured to count a network connection variable included in the scenario behavior data to obtain a first vector corresponding to the network connection variable.

[0019] The second encoding module is configured to encode a category variable included in the scenario behavior data according to a one-hot encoding mode to obtain a second vector corresponding to the category variable.

[0020] The feature splicing module is configured to splice the first vector and the second vector to obtain a feature vector corresponding to the scenario behavior data.

[0021] In an example embodiment of the present disclosure, based on the foregoing embodiment, the second encoding module can be configured to encode a category variable having two values according to a one-hot encoding mode to obtain the corresponding second vector, and perform feature induction processing on a category variable having more than three values, and perform one-hot encoding on the processed values to obtain the corresponding second vector.

[0022] In an example embodiment of the present disclosure, based on the foregoing embodiment, the apparatus can further include a missing value encoding module, which can be configured to encode the scenario behavior data according to the values of each variable contained in the scenario behavior data, to determine the feature vector.

[0023] In an example embodiment of the present disclosure, based on the foregoing embodiment, the model application module can include a training sample acquisition module, a willingness label determination module, and a model training module.

[0024] The training sample acquisition module is configured to concatenate the user portrait information and the feature vector to obtain a training sample.

[0025] The willingness label determination module is configured to determine a willingness label corresponding to the training sample according to the scenario behavior data.

[0026] The model training module is configured to train a deep learning model using the training sample and the corresponding willingness label, to obtain a trained user willingness recognition model.

[0027] In an example embodiment of the present disclosure, based on the foregoing embodiment, the target business scenario includes a second-class e-commerce advertisement, and the model application module can be configured to input user behavior data of a to-be-identified user terminal into the user willingness recognition model, to predict a probability of the to-be-identified user terminal clicking the second-class e-commerce advertisement through the user willingness recognition model.

[0028] In an example embodiment of the present disclosure, based on the foregoing embodiment, the apparatus further includes a business scenario updating module and a model updating module.

[0029] The business scenario updating module is configured to update the scenario behavior data corresponding to a changed scenario when the target business scenario changes.

[0030] The model updating module is configured to reconstruct a willingness recognition model using the user portrait information and the updated scenario behavior data, to identify an intention of the to-be-identified user terminal to the changed scenario.

[0031] According to a third aspect of the present disclosure, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the user willingness recognition method described in any of the above embodiments is implemented.

[0032] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to perform the user intention recognition method according to any of the above embodiments by executing the executable instructions.

[0033] According to a fifth aspect of the present disclosure, a computer program product or computer program is provided, comprising computer instructions stored in a computer readable storage medium. A processor of a computer device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to cause the computer device to perform the user intention recognition method provided in the above embodiments.

[0034] The exemplary embodiments of the present disclosure can have the following partial or all beneficial effects:

[0035] In the user intention recognition method provided by the example embodiment of the present disclosure, user portrait information is constructed through user behavior data of a user end, and scene behavior data is encoded according to a variable corresponding feature encoding mode to determine a feature vector corresponding to the scene behavior data. Further, a deep learning algorithm is used to train a model in combination with the user portrait information and the feature vector to obtain a trained user intention recognition model, so as to predict the intention of the user end to a target business scene through the obtained user intention recognition model.

[0036] It can be seen that, in the technical solution, the target population is not determined by human-made rules, but the user intention recognition model obtained through deep learning is used to recognize the user intention, which can improve the accuracy of intention recognition. Moreover, the user portrait information is constructed through user behavior data, which objectively and accurately represents the characteristics of the population, and the characteristics of the business scene are represented through scene behavior data, so that the model can capture high-dimensional feature information of the interaction between the user end and the target business scene, and improve the model precision. At the same time, the target business scene and the user behavior data of the user end itself are combined, which can make the user intention recognition model meet the recognition needs of different scenes, improve the reusability of the model, and reduce the cost.

[0037] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF DRAWINGS

[0038] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the application. It is apparent that the accompanying drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0039] Figure 1 A system architecture diagram of a user intention recognition method to which an embodiment of the present disclosure can be applied is shown.

[0040] Figure 2 A flowchart diagram of a user intention recognition method according to an embodiment of the present disclosure is shown schematically.

[0041] Figure 3 A flowchart diagram of a user intention recognition method according to another embodiment of the present disclosure is shown schematically.

[0042] Figure 4 A flowchart diagram of a user intention recognition method according to an embodiment of the present disclosure is shown schematically.

[0043] Figure 5 A flowchart diagram of a user intention recognition method according to another embodiment of the present disclosure is shown schematically.

[0044] Figure 6 A test effect diagram of a user intention recognition model in an embodiment of the present disclosure is shown schematically.

[0045] Figure 7 An application effect diagram of a user intention recognition model according to an embodiment of the present disclosure is shown schematically.

[0046] Figure 8 A flowchart diagram of a user intention recognition method according to an embodiment of the present disclosure is shown schematically.

[0047] Figure 9 A structural diagram of a user intention recognition device to which an embodiment of the present disclosure can be applied is shown.

[0048] Figure 10 A structural diagram of a computer system of an electronic device suitable for implementing an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0049] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations can be implemented in any number of ways, and example implementations should not be construed as limited to only those described herein; rather, embodiments should be construed more broadly. It will be appreciated that aspects of the present disclosure can be implemented in various forms and that the present disclosure should not be construed as limited to only the examples set forth herein; rather, examples are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example implementations to those skilled in the art. Described features, structures, or characteristics can be combined in any suitable manner in one or more implementations. In the following description, numerous specific details are provided to give a thorough understanding of implementations of the present disclosure. One skilled in the relevant art will recognize, however, that the aspects of the present disclosure can be practiced without one or more of the specific details, or

[0050] Furthermore, the accompanying drawings are only schematic and are non-limiting detailed representations of examples of the present disclosure. Like references numerals can be used to denote like parts throughout the specification and / or drawings and a repetitive description is omitted concomitantly. Some of the blocks in the drawings can be functional blocks that do not necessarily have a corresponding physical structure or implementation. These blocks can be implemented by software or firmware or in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0051] In the technical solution, the user terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, or the like, but is not limited thereto. The user terminal and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in the present application.

[0052] The use scenario of the user willingness recognition method provided by the technical solution can be that a second-class e-commerce usually takes an H5 landing page as a basis, reaches users in the form of information flow advertising, and takes a cash on delivery mode as a promotion mode. When a second-class e-commerce advertisement is precisely placed, users with high willingness to the second-class e-commerce advertisement need to be identified.

[0053] The use scenario of the user willingness recognition method provided by the technical solution can also be that with the development of AI (Artificial Intelligence) technology, machine learning models are applied more and more widely. Then, when a user browses Internet information, the user's willingness is identified, and the user is recommended the needed information according to the user's willingness.

[0054] Artificial Intelligence (AI) is the theory, method, technology and application system of using digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which tries to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0055] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction system, mechatronics, etc. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning, etc.

[0056] Machine Learning (ML) is a multi-disciplinary subject, involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory, etc. It is a subject that studies how computers simulate or implement human learning behavior to acquire new knowledge or skills, and reorganize existing knowledge structure to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent, and its applications are widespread in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and example-based learning.

[0057] With the research and progress of artificial intelligence technology, artificial intelligence technology has been researched and applied in many fields, such as common smart home, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned vehicles, autonomous vehicles, drones, robots, smart medical care, smart customer service, etc. It is believed that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0058] Specifically, Figure 1 A schematic diagram of a system architecture showing an exemplary application environment of a user intention recognition method to which embodiments of the present disclosure can be applied is shown.

[0059] As Figure 1 shown, the system architecture 100 can include one or more of user end devices 101, 102, 103 as user end, a network 104, and a server 105.

[0060] The user terminal devices 101, 102, and 103 can be various electronic devices with display screens, including but not limited to desktop computers, portable computers, smart phones, and tablet computers, and the like.

[0061] The network 104 is a medium for providing a communication link between the user terminal devices 101, 102, and 103 and the server. The network 104 can include various connection types, such as wired, wireless communication links, or fiber optic cables, and the like.

[0062] The server 105 can be 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, security services, and big data and artificial intelligence platforms.

[0063] It should be understood that Figure 1 The number of user terminal devices, networks, and servers in the above system architecture is only illustrative. Any number of user terminal devices, networks, and servers can be provided as needed. For example, the server 105 can be a server cluster composed of multiple servers.

[0064] It should be noted that in the specific embodiments of the present application, when the embodiments of the present application are applied to specific products or technologies, user permission or consent is required, and the collection, use, and processing of related data need to comply with relevant laws, regulations, and standards of the country and region

[0065] Based on the above system architecture, the embodiments of the user intention recognition method provided by the present technical solution are described in detail as follows:

[0066] Figure 2 A flowchart of a user intention recognition method according to an embodiment of the present disclosure is schematically shown. Specifically, the user intention recognition method provided by the present embodiment takes the server as the execution subject. Referring to Figure 2 The embodiment shown in the figure includes steps S21, S22, and S23.

[0067] In step S21, user portrait information is constructed through user behavior data of the user terminal.

[0068] The user behavior data can include identification information of the user, such as a user name, a user ID, and the like, and can also include identification information and attribute information of the user terminal itself, such as a device ID, a device model, and the like, or network connection information of the user terminal, such as a network type, a connection duration, and the like. The present embodiment does not limit this.

[0069] Each service request of the user when the user terminal is online is recorded through online log data, so that real-time recorded log data can be pulled from the online cache, and the pulled data is stored in a distributed file system (Hadoop Distributed File System, which can be referred to as HDFS). The HDFS has high fault tolerance and can be designed for deployment on low-cost hardware, providing high throughput for accessing data and being suitable for scenarios of super-large-scale data sets. The log data can be read from the HDFS, and the required data is extracted from the read log data as user behavior data. Alternatively, the log data is calculated and processed offline, so as to obtain the required user behavior data. For example, the user ID, device model, network connection type, network connection time, and the like recorded by the user terminal each time the user terminal is online are extracted from the log data as user behavior data.

[0070] Analyzing and calculating all user behavior data of each user terminal can construct user portrait information of the user terminal. The user portrait information can include basic attribute features of the user terminal and online time features of the user terminal. The basic attribute features can include user attributes such as user gender and user age, device attributes such as device type and device level of the user terminal, and network connection attributes such as network type (such as hot spot connection and broadband connection), connection time, and disconnection time. The online time features can refer to the time regularity of the user terminal online, and can specifically include the time period of the user terminal online, such as weekend, weekday time period, active time, and rest time.

[0071] The user behavior data pulled from the user terminal can analyze which time period the user terminal is online and the corresponding network connection attribute of the time period, for example, the user terminal is online in the form of hot spot connection from 20:00 to 22:00 on weekdays and in the form of broadband connection from 11:00 to 14:00 on weekends. In an example embodiment, the user behavior data of the user terminal in different time windows is counted, and the online time features and the basic attribute features of the user terminal are obtained as the user portrait information. Because the online behaviors of the user in different periods are different, the data accumulation in different time windows can better reflect the behavior characteristics of the user. For example, the user behavior data of the user terminal in the past half year, the past three months, the past month, the past week, and the past three days is counted offline at regular intervals, so that the user portrait information in different time windows is obtained, such as the network connection type of the user in the past half year and the corresponding online time period.

[0072] In step S22, the scene behavior data of the user terminal in the target business scenario is collected, and the variables included in the scene behavior data are encoded according to the feature encoding mode corresponding to the variables to determine the feature vector corresponding to the scene behavior data.

[0073] The scenario behavior data is data recorded by a user in a target business scenario, which can include multiple variables, and the values of the variables included in each piece of scenario behavior data are different. Specifically, the scenario behavior data can include advertising information, such as advertising attributes, advertising categories, advertising channels, and the like, and can also include information of a user side and behavior information of the user side to the advertising, such as a current online time of the user side, a browsing time of the advertising, whether to click the advertising, whether to purchase an advertising product, and the like. The scenario behavior data corresponding to the target business scenario can also be extracted from log data recorded by the user side. Specifically, the log data of the target business scenario is usually saved in a specific directory corresponding to the business scenario, and data is pulled from the specific directory to a server or stored in a file storage system by a data collection tool at a regular time.

[0074] For example, the scenario behavior data is updated according to a certain time period, such as collected once a week, updated once every 10 days, updated once a month, updated once every three months, and the like, so that the scenario behavior data remains a certain timeliness, and the accuracy of model recognition is improved.

[0075] Each variable in the scenario behavior data can correspond to a different feature encoding mode. In the scenario of two-class e-commerce advertising, the scenario behavior data of a user is relatively complex, and feature information is difficult to display and express on a data representation, which leads to that deep learning cannot be used to learn and process data in this scenario. Therefore, in the embodiment, multiple encoding modes are used to encode data according to variables included in the scenario behavior data, and each variable can adopt a different feature encoding mode, so as to fully represent the features and improve the accuracy of feature representation, so as to facilitate the machine learning model to process the data.

[0076] Encoding data means that data that cannot be recognized by a computer is encoded into data that can be recognized and processed by the computer, that is, a feature is converted into a digital representation. For example, gender “male, female” cannot be processed in calculation, which leads to the loss of the feature. “Male” can be encoded as 1, and “female” can be encoded as 0, so that calculation and other data processing can be performed, and the model learns the relationship between the feature and other features.

[0077] In this embodiment, all possible variables included in the scene behavior data are analyzed in advance to determine all variable types, such as gender variable, network connection variable, advertisement category variable, device category variable, application program category, numerical variable, and the like. A correspondence between each variable type and a feature encoding mode is established, so that each variable can correspond to one feature encoding mode, for example, the gender variable adopts the one-hot encoding technology. In addition, different variables according to actual requirements can also correspond to the same feature encoding mode, for example, the gender variable corresponds to the one-hot encoding feature encoding mode, and the network connection variable also adopts the one-hot encoding for encoding.

[0078] Exemplarily, Figure 3 A flowchart of an encoding method of an embodiment of the present disclosure is schematically shown. The execution subject of this embodiment can be a server. In this embodiment, the scene behavior data can include network connection variables and category variables. As shown in FIG. 3, this embodiment can include steps S31, S32, and S33. Figure 3

[0079] In step S31, the network connection variable included in the scene behavior data is counted to obtain a first vector of the network connection variable. The network connection variable included in the scene behavior data is counted, and the obtained counting result is taken as the first vector corresponding to the network connection variable. Generally, different network connection modes are adopted by user terminals in different places, so the value of the network connection variable can reflect the place characteristics of the user terminal. The network connection variable can include multiple values, and the value of the network connection variable can be the name of the connected network, such as the name of a WiFi hotspot, the name of a broadband, and the like. Each value of the network connection variable is counted to obtain a counting result, which is taken as the encoding of the variable, so as to convert the variable into a vector. For example, the scene behavior data within half a month is collected, the network connection variable is the hotspot of “food-Chinese food-Cantonese food” for 3 times, then the encoding of “food-Chinese food-Cantonese food” is 3, the network connection variable is the hotspot of “drinks-tea” for 5 times, then the encoding corresponding to “drinks-tea” is 5, and the first vector corresponding to the network connection variable can be obtained as (3, 5).

[0080] ​In step S32, the category variable contained in the scene behavior data is encoded in a one-hot encoding manner to obtain a second vector corresponding to the category variable. One-hot encoding is also known as one-hot encoding or one-bit effective encoding in the art, which mainly uses an N-bit state register to encode N states, each state has its own register bit, and only one bit is valid at any time. Using one-hot encoding can expand the value of discrete features to Euclidean space, and a certain value of a discrete feature corresponds to a point in Euclidean space. The category variable is data in the scene behavior data that has a classification category, and its value can be a specific classification category, such as the category of an advertising product "daily necessities-dinnerware-ceramic products". According to all values of the category variable, a corresponding encoding value is determined for each value, for example, the category variable has three values, and after one-hot encoding, the three values can correspond to 100, 010, and 001, respectively. According to the specific value of the category variable in the scene behavior data, the second vector corresponding to the variable can be determined, for example, the value of the category variable is "food-biscuit", which corresponds to the encoding 100, and the corresponding second vector is 100.

[0081] Exemplary, Figure 4 A flowchart of a method for obtaining a second vector for a category variable according to one embodiment of the present disclosure is shown schematically. Referring to Figure 4 The method can include:

[0082] In step S41, the category variable with two values is encoded in a one-hot encoding manner to obtain a corresponding second vector. In step S42, the category variable with more than three values is processed by feature induction, and the processed values are one-hot encoded to obtain a corresponding second vector.

[0083] The scene behavior data can include multiple category variables, such as system version category, advertising product category, gender classification, network connection classification, etc. Different category variables have different values, and when a category variable has only two values, one-hot encoding can be used to encode to obtain a second vector. When a category variable has multiple values, the values of the category variable can be processed by feature induction to reduce the values of the category variable. For example, the multiple values of the system version feature of an Android phone include "4.2", "4.4", and "5.0", etc., which can be processed by feature induction as "low version Android system", and values greater than "5.0" such as "5.1, 6.0" can be processed as "high version Android system". Experiments show that the feature induction processing method can bring greater positive benefits than one-hot encoding of each value of the "Android system version" category variable.

[0084] In step S33, the first vector and the second vector are spliced to obtain a feature vector corresponding to the scene behavior data. It should be understood that the description of the first vector and the second vector is to facilitate the description of different feature encoding modes corresponding to the network connection variable and the category variable. In fact, each variable in the scene behavior data can be converted into a vector, and finally the vectors corresponding to the variables are connected to obtain a complete feature vector corresponding to the scene behavior data. In addition, in the embodiments of the present disclosure, the scene behavior data further includes other types of variables, such as numerical variables, missing value variables, etc.

[0085] In the example embodiment, the network connection variable can represent the trajectory of the user's network connection, such as from broadband connection to WiFi hotspot connection, from home hotspot to company hotspot, etc., so the embodiment can also encode the trajectory of the network connection variable to capture the information of the user's network behavior. For example, the trajectory feature of the network connection is extracted by a convolutional neural network based on a minimum spanning tree (MST), and the trajectory is converted into an Embedding feature.

[0086] In the example embodiment, the strong sparsity of the category variable can cause model overfitting. In order to avoid the overfitting and improve the stability of the model, for the category variable, a neural network Embedding can also be used to convert the high-dimensional sparse classification category variable into a low-dimensional dense Embedding variable. Embedding can convert discrete variables into continuous vector representation. In addition, the List-Embedding based method can also extract Embedding of different values of the category variable, such as extracting Embedding of the traffic usage of different category APPs, etc., to obtain a low-dimensional dense feature representation.

[0087] In the example embodiment, for the numerical variable, the normalization method can be used to eliminate the dimensional difference between the features, so that the model is more stable. For example, for the features conforming or approximately conforming to the normal distribution, the Gaussian normalization method is selected.

[0088] Data often exist missing values, for the scene behavior data in the variable can be filled with missing values, for example, according to a specific value or average value filling, etc., or directly to the missing value rejection; Or according to the missing mark way to deal with missing values, etc., in order to convert the scene behavior data into feature vector. But, the missing value directly filling or rejection and so on processing, the corresponding variable for model learning will reduce the meaning. Therefore, according to the example, according to the value of each variable contained in the scene behavior data, the scene behavior data is encoded with missing values to determine the feature vector. Among them, the missing value encoding can refer to the encoding of the variable value state of the scene behavior data, so as to represent the integrity of the data. For example, the variable with value can be encoded as 1, and the variable with missing value can be encoded as 0, so as to convert the missing value state of the scene behavior data into a vector, for example, the scene behavior data contains 5 variables, and the value of the first variable is missing, and the corresponding encoding result is 01111. It can be seen that the embodiment can increase a feature representing the missing state of the data by encoding the missing value, so as to make full use of the missing value to bring the maximum positive benefit to the model.

[0089] After encoding the missing values of the scene behavior data, the encoding result can be used as a feature, and the vectors corresponding to each variable are connected to obtain the feature vector of the scene behavior data.

[0090] Each kind of feature encoding mode can obtain a feature (vector) of the scene behavior data, and the obtained features are spliced to finally obtain the feature vector of the scene behavior data. For example, the network connection variable in the scene behavior data is first encoded by counting to obtain a first vector; the category variable in the scene behavior data is then encoded by one-hot encoding to obtain a second vector; the category variable in the scene behavior data is then processed by Embedding to obtain a low-dimensional third vector; the scene behavior data is then processed by missing value encoding to obtain a fourth vector; and the first vector, the second vector, the third vector and the fourth vector are combined to obtain the final feature vector of the scene behavior data. For another example, the values corresponding to each category variable are processed by Embedding to obtain a first vector; the values of the category variable are then extracted by Embedding to obtain a second vector; the numerical type variable in the scene behavior data is then processed by normalization to obtain a third vector; the trajectory of the network connection variable is then embedded by the MST-CNN deep learning network to obtain a fourth vector; and the first, second, third and fourth vectors are combined to obtain the final feature vector. It can be seen that the embodiment can represent the features from various dimensions, so as to make the model learn the interaction between the features and improve the recognition accuracy of the model.

[0091] In other embodiments of the present disclosure, the feature encoding manner can also include other manners, for example, Label Encoding, which is a manner of converting a feature variable into a digital representation by customizing a digital label for the feature variable; or Dummy coding, and the like. The embodiments of the present disclosure are not limited thereto.

[0092] With reference to Figure 2 In step S23, the user portrait information and the feature vector are trained by using a deep learning algorithm to obtain a trained user willingness recognition model, so as to predict the willingness of a to-be-identified user terminal to the target business scenario by using the user willingness recognition model.

[0093] In the present embodiment, the training sample can be obtained by combining the user portrait information and the feature vector corresponding to the scenario behavior data, so as to train the deep learning model by using the training sample. The specific process of training can be as shown in Figure 5 With reference to Figure 5 The method for training the user willingness recognition model can include steps S51, S52 and S53.

[0094] In step S51, the user portrait information and the feature vector are spliced to obtain a training sample of the user willingness recognition model. For each user terminal, the user portrait information of the user terminal and the feature vector corresponding to the scenario behavior data of the user terminal are spliced, so as to combine the features of the user behavior data and the scenario behavior data to obtain the training sample corresponding to the user terminal. The training sample combines the features of the user behavior data and the scenario behavior data, and can represent the willingness of the user terminal in the target business scenario.

[0095] In step S52, the willingness label corresponding to the training sample is determined according to the scenario behavior data. Specifically, the scenario behavior data can include a specific variable, which can be used to represent the willingness of the user to the target business scenario, such as a click behavior or a purchase behavior of an advertisement. For example, if the value of the specific variable in the scenario behavior data is 1, it can be determined that the willingness label is 1, indicating that the user terminal has the willingness, and if the value of the specific variable is 0, it can be determined that the user terminal does not have the willingness, and the willingness label is 0. The training sample can be divided into positive samples and negative samples by using the willingness label.

[0096] In step S53, the deep learning model is trained by using the training samples and the corresponding willingness labels to obtain a trained user willingness recognition model. The deep learning model can be constructed by various deep learning algorithms, such as multilayer perceptron and neural network, deep neural network, recurrent neural network, and recursive neural network. By constructing the deep learning model, the feature learning of the training samples can be performed, and the mapping rule between the training samples and the willingness labels thereof can be determined.

[0097] For example, a deep learning model RALM (real-time attention based look-alike model) can be used for training, and a user willingness recognition model meeting the requirements can be obtained after the training is completed.

[0098] The RALM model is essentially a similarity-based model, which includes two parts of “user representation learning” and “look-alike model learning”. The look-alike model adopts a double-tower structure, the input on the left side is the Embedding vector of the seed user, and the input on the right side is the Embedding vector of the target user (to-be-recognized user end). The Embeddings on both sides are mapped to a low-dimensional space through a fully connected layer. The seed user can refer to the user end corresponding to the training sample, and the target user refers to the to-be-recognized user end. Since the Embedding vector of the target user on the right side is obtained through user representation learning, in order to prevent overfitting, the first fully connected layer of the double tower is shared. After the fully connected layer, the tower on the left side can obtain the Embedding corresponding to each cluster, and then the Embedding of each cluster and the Embedding of the target user are respectively input into the Global Attention Unit (global attention unit) and the Local Attention Unit (local attention unit) to obtain the Global Embedding (global feature representation) and the Local Embedding (local feature representation). This model optimizes the effect of user representation learning, introduces a deep interest network of “attention fusion layer” for multi-domain user interest representation learning, can solve the overfitting and noise problems caused by strong and weak related features respectively, and the attention fusion layer can capture the interest preferences of users in various scenarios. The global attention unit can weight the representation of a single user and penalize noise users, which is more robust than using the same weight for all users. The local attention unit learns the local representation of the seed user, which can weight the relevance between the seed user and the target user. For different target users, the learned seed user representation is different, which can greatly improve the expression ability of the seed user.

[0099] In the model training stage, the FALM model extracts features through clustering. The clustering process iterates repeatedly and is time-consuming. The number of clustering centers affects the clustering effect. The number of clustering centers can be flexibly set according to actual requirements, such as 10, 20, 30, etc. In the business scenario of two-class e-commerce advertising, the number of clustering centers can be selected as 10-50, that is, the clustering effect can be ensured and the time consumption is reduced to a certain extent.

[0100] In the data processing stage of encoding the scene behavior data to obtain the feature vector, and the model training stage, different computing engines can be used to provide computing support in the exemplary embodiments. For example, for non-Embedding features, a spark computing engine can be used for computing processing, and for Embedding type features extracted through Embedding, a TensorFlow computing engine can be used for computing, and the results are stored in HDFS, so as to determine the feature vector of the scene behavior data. When training, the computing results can be read from HDFS, and then a model is built based on TensorFlow to obtain the user intention recognition model.

[0101] After training, the obtained user intention recognition model can be evaluated. If the evaluation index meets the requirements, the model can be pushed online for A / B Test to further determine the test index of A / B Test. If the test index is normal, the model can be used formally with all traffic. If the evaluation index does not meet the requirements, the model can be retrained until the model effect meets the requirements. A / B Test refers to an online testing method used to test new products or new functions. Online traffic can be divided into two groups, one group uses new products or new functions as the experimental group, and the other group uses existing products or functions as the control group, so as to determine whether the new products or new functions bring better effect. For example, in the business scenario of two-class e-commerce advertising, the user terminal is recognized by using the user intention recognition model in the embodiment, and the intention of the user terminal is recognized by using the original method. The recognition effect of the test can be as shown in Figure 6 For example, the AUC value is used as the evaluation index of the model. The larger the AUC value, the better the recognition effect of the model. Figure 6 The offline AUC can represent the model recognition effect when the user intention recognition model is tested by using test data in an offline manner. The online AUC is the recognition effect of the user intention recognition model for recognizing the online user terminal to be recognized. It can be seen that the user intention recognition model based on the RALM algorithm has higher recognition effect.

[0102] With the business scenario of the second type of e-commerce advertisement as an example, after the user willingness recognition model is trained, the user willingness recognition model can be used to recognize the probability of the online to-be-recognized user end clicking the second type of e-commerce advertisement. Specifically, the user behavior data of the to-be-recognized user in a certain time period can be collected first, the collected user behavior data is input into the user willingness recognition model to obtain a prediction result, and according to the prediction result, it can be determined whether the to-be-recognized user will click the second type of e-commerce advertisement. The prediction result can represent the probability of the user clicking the second type of e-commerce advertisement. According to the prediction result, the second type of e-commerce advertisement can be pushed to the corresponding user end. When the probability is greater than a certain threshold, the second type of e-commerce advertisement can be pushed to the to-be-recognized user end, for example, when the prediction result is greater than 0.6, the second type of e-commerce advertisement is recommended, and the like. According to the recommendation effect after the recommendation, the model can be evaluated again, so that when the recommendation effect is low, the model can be trained and improved again to improve the effect of the model application. For example, after the second type of e-commerce advertisement is recommended to the to-be-recognized user with a high prediction result, the feedback behavior of the user can be received, such as a click behavior of clicking the second type of e-commerce advertisement, or a purchase behavior of purchasing the goods corresponding to the second type of e-commerce advertisement, and the like. Figure 7 An application effect diagram of the user willingness recognition model applied to the recommendation of the second type of e-commerce advertisement is schematically shown. Figure 7 By artificially formulating rules, a non-deep learning scheme, and the user willingness recognition model in the embodiment, the willingness of the user end is recognized respectively, and the second type of e-commerce advertisement is recommended according to the recognition result, it can be seen that the user willingness recognition model in the embodiment can achieve a high advertisement click rate and a high purchase rate.

[0103] In the example embodiments of the present disclosure, the target business scenario can be changed according to actual needs, when the target business scenario changes, the target business scenario is updated to a changed scenario, and the scene behavior data corresponding to the changed scenario is further updated; the user willingness recognition model conforming to the changed scenario is reconstructed by using the updated scene behavior data and the user portrait information, so as to recognize the willingness of the to-be-recognized user end to the changed scenario. For example, the changed scenario can be a game application advertisement, the model is updated by collecting the scene behavior data corresponding to the game application advertisement, for example, the scene behavior data corresponding to the game application advertisement can include the information of the game, such as the game name, the game classification type, and can also include the behavior information of the user end to the game application advertisement, such as the click behavior, the download behavior, and the like. According to the updated user willingness recognition model, the probability of the to-be-recognized user end clicking the game application advertisement is recognized, and then the game application advertisement is recommended to the user end with a high probability.

[0104] Figure 8 A flowchart of the user willingness recognition method in an embodiment of the present disclosure is schematically shown. Figure 8As shown, the method can include steps S801 to S811. Among them:

[0105] In step S801, user behavior data is extracted; in step S802, user portrait information is determined; in step S803, scene behavior data corresponding to the second type of e-commerce advertisement in the past month of the user end is accumulated, and a feature vector is calculated; in step S804, the user portrait information is spliced with the feature vector; in step S805, a user willingness recognition model is trained through the spliced features; in step S806, the user willingness recognition model is evaluated offline; offline evaluation can refer to testing the model using pre-collected test data; in step S807, it is determined whether the evaluation result meets the expectation, if yes, step S808 is executed, if not, the model is returned to step S805 for continuous training; a threshold value can be set in advance, if the evaluation result exceeds the threshold value, it can be determined that the expectation is met, if the evaluation result does not exceed the threshold value, the expectation is not met, for example, an AUC value is set, if the value of the model offline test evaluation is not greater than the AUC value, it is determined that the expectation is not met, if the test value is greater than the AUC value, it can be determined that the expectation is met; in step S808, the user willingness recognition model is evaluated online; online evaluation can refer to identifying online user terminals using the model, comparing the difference between the identification result and the original online user terminal identification result, such as A / B test test method; in step S809, it is determined whether the online effect meets the expectation, if yes, step S810 is executed, if not, the model is returned to step S805 for continuous training; the online effect can be judged by using click rate, purchase rate and other indicators, for example, if the click rate exceeds a certain value, it is determined that the expectation is met, if the click rate does not exceed the certain value, the expectation is not met; in step S810, the user willingness recognition model is called in the online service, and the second type of e-commerce advertisement is recommended to the user; in step S811, the feedback behavior of the user terminal after the recommendation is obtained; the feedback behavior can be used to further determine the effect of the user willingness recognition model in the online service, facilitating subsequent adjustment of the model. It can be seen that the user behavior data of the user terminal and the business scenario of the second type of e-commerce advertisement can be combined, so that the user willingness recognition model based on deep learning can be adapted to the second type of e-commerce advertisement scenario, the recognition accuracy of the second type of e-commerce advertisement of the user terminal is improved, and the second type of e-commerce advertisement is accurately recommended.

[0106] A person skilled in the art can understand that all or part of the steps of the above embodiments are implemented as computer programs executed by a processor (including CPU and GPU). When the computer program is executed by the processor, the above functions defined by the above method provided by the present disclosure are executed. The program can be stored in a computer readable storage medium, which can be a read-only memory, a disk or an optical disk, etc.

[0107] In addition, it should be noted that the above-described figures are only schematic representations of the processes included in the method according to the example embodiment of the present disclosure, and are not intended to be limiting. It will be readily appreciated that the processes shown in the above-described figures do not indicate or limit the chronological order of these processes. In addition, it will also be readily appreciated that these processes can be executed synchronously or asynchronously, for example, in a plurality of modules.

[0108] The user intention recognition device provided in the present technical solution is described below:

[0109] The user intention recognition device provided in the present example embodiment can be used to execute the user intention recognition method described above in the present disclosure. Referring to Figure 9 As shown in the figure, the user intention recognition device 90 includes a user portrait module 91, a feature representation module 92, and a model application module 93.

[0110] The user portrait module 91 is configured to construct user portrait information based on user behavior data of a user terminal.

[0111] The feature representation module 92 is configured to collect scenario behavior data corresponding to a target business scenario, encode variables included in the scenario behavior data according to a variable corresponding feature encoding mode, and determine a feature vector corresponding to the scenario behavior data.

[0112] The model application module 93 is configured to train the user portrait information and the feature vector using a deep learning algorithm, obtain a trained user intention recognition model, and predict the intention of a user terminal to be identified to the target business scenario through the user intention recognition model.

[0113] In an example embodiment of the present disclosure, based on the foregoing embodiment, the user portrait module 91 can be configured to count user behavior data of the user terminal in different time windows to obtain online time features and basic attribute features of the user terminal as the user portrait information.

[0114] In an example embodiment of the present disclosure, based on the foregoing embodiment, the feature representation module 92 can include a first encoding module, a second encoding module, and a feature splicing module.

[0115] The first encoding module can be configured to count network connection variables included in the scenario behavior data to obtain a first vector corresponding to the network connection variables.

[0116] The second encoding module is configured to encode category variables included in the scenario behavior data according to a one-hot encoding mode to obtain a second vector corresponding to the category variables.

[0117] The feature splicing module is configured to splice the first vector and the second vector to obtain a feature vector corresponding to the scenario behavior data.

[0118] In an example embodiment of the present disclosure, based on the foregoing embodiment, the second encoding module can be configured to encode a category variable having two values in a one-hot encoding manner to obtain the corresponding second vector, and to perform feature induction processing on a category variable having more than three values and perform one-hot encoding on the processed values to obtain the corresponding second vector.

[0119] In an example embodiment of the present disclosure, based on the foregoing embodiment, the device 90 can further include a missing value encoding module configured to encode the scenario behavior data according to the values of each variable included in the scenario behavior data to determine the feature vector.

[0120] In an example embodiment of the present disclosure, based on the foregoing embodiment, the model application module 93 can include a training sample acquisition module, a willingness label determination module, and a model training module.

[0121] The training sample acquisition module is configured to splice the user portrait information and the feature vector to obtain a training sample.

[0122] The willingness label determination module is configured to determine a willingness label corresponding to the training sample according to the scenario behavior data.

[0123] The model training module is configured to train a deep learning model using the training sample and the corresponding willingness label to obtain a trained user willingness recognition model.

[0124] In an example embodiment of the present disclosure, based on the foregoing embodiment, the target business scenario includes a second type of e-commerce advertisement, and the model application module 93 can be configured to input user behavior data of a user terminal to be recognized into the user willingness recognition model to predict a probability of the user terminal to be recognized clicking on the second type of e-commerce advertisement through the user willingness recognition model.

[0125] In an example embodiment of the present disclosure, based on the foregoing embodiment, the device 90 further includes a business scenario updating module and a model updating module.

[0126] The business scenario updating module is configured to update the scenario behavior data corresponding to a changed scenario when the target business scenario changes.

[0127] The model updating module is configured to reconstruct a willingness recognition model using the user portrait information and the updated scenario behavior data to recognize an intention of the user terminal to be recognized to the changed scenario.

[0128] The specific details of each module or unit in the above user intention recognition apparatus have been described in detail in the corresponding user intention recognition method, and thus will not be described here again.

[0129] Figure 10 A structural diagram of a computer system of an electronic device suitable for implementing embodiments of the present application is shown.

[0130] It should be noted that, Figure 10 The computer system 1000 of the electronic device shown is only an example and should not impose any limitation on the functions and use range of embodiments of the present application.

[0131] As Figure 10 shown, the computer system 1000 includes a processor 1001, which can include a Graphics Processing Unit (GPU), a Central Processing Unit (CPU), which can perform various appropriate actions and processes according to programs stored in a Read-Only Memory (ROM) 1002 or programs loaded from a storage portion 1008 into a Random Access Memory (RAM) 1003. Various programs and data required for system operation are also stored in the RAM 1003. The processor (GPU / CPU) 1001, the ROM 1002, and the RAM 1003 are connected to each other through a bus 1004. An Input / Output (I / O) interface 1005 is also connected to the bus 1004.

[0132] The following components are connected to the I / O interface 1005: an input portion 1006 including a keyboard, a mouse, and the like; an output portion 1007 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and the like, and a speaker, and the like; a storage portion 1008 including a hard disk, and the like; and a communication portion 1009 including a network interface card such as a LAN (Local Area Network) card, a modem, and the like. The communication portion 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as necessary. A removable medium 1011 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, and the like is mounted on the drive 1010 as necessary, so that a computer program read therefrom is installed in the storage portion 1008 as necessary.

[0133] In particular, according to embodiments of the present disclosure, the processes described below with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product including a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network by the communication section 1009, and / or installed from the detachable medium 1011. When the computer program is executed by the processor (GPU / CPU) 1001, various functions defined in the system of the present application are executed. In some embodiments, the computer system 1000 can also include an AI (Artificial Intelligence) processor for processing computation operations related to machine learning.

[0134] It should be noted that the computer-readable medium shown in the embodiments of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may, for example, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or apparatus, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (Compact Disc Read-Only Memory, CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, device or apparatus. In the present disclosure, the computer-readable signal medium can include a data signal carried in a baseband or as a part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take many forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or apparatus. The program code contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to wireless, wired, or the like, or any suitable combination of the above.

[0135] The flowcharts and block diagrams in the drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowcharts or block diagrams can represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks noted in succession can in fact be executed substantially concurrently or the blocks can sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowcharts, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0136] The units described in the embodiments of the present disclosure can be implemented by software, or by hardware, or be implemented by a combination of software and hardware. The names of the units described above do not constitute a limitation on the units themselves in some cases.

[0137] As another aspect, the present disclosure also provides a computer readable medium, which can be included in the electronic device described in the above embodiments, or can exist separately without being assembled into the electronic device. The computer readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to implement the method described in the above embodiments.

[0138] For example, the electronic device can implement the steps shown in Figure 2 S21, constructing user portrait information through user behavior data of a user terminal; S22, collecting scenario behavior data of the user terminal in a target business scenario, encoding variables included in the scenario behavior data according to a feature encoding mode corresponding to the variables, to determine a feature vector corresponding to the scenario behavior data; and S23, training the user portrait information and the feature vector by using a deep learning algorithm, to obtain a trained user willingness recognition model, so as to predict the willingness of a to-be-identified user terminal to the target business scenario through the user willingness recognition model.

[0139] For another example, the electronic device can implement the steps shown in Figures 3 to 5

[0140] ​It should be noted that, although several modules or units of the devices for action execution are mentioned in the above detailed description, such division is not mandatory. Indeed, according to embodiments of the present disclosure, features and functionalities of two or more modules or units described above can be embodied in one module or unit. Conversely, features and functionalities of one module or unit described above can be further divided into multiple modules or units.

[0141] From the above description of the embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product. The software product can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash disk, a mobile hard disk, or the like) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to perform the methods according to the embodiments of the present disclosure.

[0142] Other embodiments of the present disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the application disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure following the general principles thereof and including such departures from the present disclosure that come within known use or custom in the art. The specification and examples are to be regarded as illustrative only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0143] It should be understood that the present disclosure is not limited to the precise structures described and shown in the drawings, and that various modifications and changes can be made without departing from the scope thereof. The scope of the present disclosure is limited only by the claims that follow.

Claims

1. A method for recognizing user intent, characterized in that, include: User profile information is constructed by using user behavior data from the user's end. The user profile information includes online time characteristics and basic attribute characteristics including network connection attributes. Collect scenario behavior data of the user terminal in the target business scenario, and encode the variables contained in the scenario behavior data according to the feature encoding method corresponding to the variables to determine the feature vector corresponding to the scenario behavior data; The user profile information and the feature vector are trained using a deep learning algorithm to obtain a trained user intention recognition model, so as to predict the intention of the user to be identified for the target business scenario through the user intention recognition model; The variables include network connection variables and category variables, which are encoded according to the feature encoding method corresponding to the variables to determine the feature vector corresponding to the scene behavior data, including: counting the values ​​of the network connection variables contained in the scene behavior data to obtain the first vector of the network connection variables; The category variables contained in the scene behavior data are encoded using a one-hot encoding method to obtain the second vector corresponding to the category variables; the first vector and the second vector are concatenated to obtain the feature vector corresponding to the scene behavior data.

2. The user intention recognition method according to claim 1, characterized in that, Building user profiles based on user behavior data from the user's end includes: The user behavior data of the user terminal in different time windows is statistically analyzed to obtain the online time characteristics and basic attribute characteristics of the user terminal, which are used as the user profile information.

3. The user intention recognition method according to claim 1, characterized in that, The category variables contained in the scene behavior data are encoded using a one-hot encoding method to obtain the second vector corresponding to the category variables, including: The category variable that has two possible values ​​is encoded using a one-hot encoding method to obtain the corresponding second vector; Category variables with three or more possible values ​​are subjected to feature summarization processing, and the processed values ​​are one-hot encoded to obtain the corresponding second vector.

4. The user intention recognition method according to claim 1, characterized in that, The method further includes: The scene behavior data is encoded with missing values ​​according to the values ​​of each variable contained in the scene behavior data in order to determine the feature vector.

5. The user intention recognition method according to claim 1, characterized in that, The user intention recognition model is obtained by training the user profile information and the feature vector using a deep learning algorithm, including: The user profile information and the feature vector are concatenated to obtain training samples; Determine the intention label corresponding to the training sample based on the scenario behavior data; The deep learning model is trained using the training samples and corresponding intention labels to obtain a trained user intention recognition model.

6. The user intention recognition method according to claim 1, characterized in that, The target business scenario includes two types of e-commerce advertising, and the step of predicting the user's intention towards the target business scenario through the user intention recognition model includes: User behavior data of the user terminal to be identified is input into the user intention recognition model to predict the probability of the user terminal to be identified clicking on the second type of e-commerce advertisement.

7. The user intention recognition method according to claim 1, characterized in that, The method further includes: When the target business scenario changes, the scenario behavior data corresponding to the changed scenario is updated; The intention recognition model is reconstructed using the user profile information and the updated scene behavior data to identify the intention of the user to be identified in the changed scene.

8. A user intention recognition device, characterized in that, include: The user profiling module is used to construct user profile information based on user behavior data from the user terminal. The user profile information includes online time characteristics and basic attribute characteristics including network connection attributes. The feature representation module is used to collect scene behavior data corresponding to the target business scenario, and encode the variables contained in the scene behavior data according to the variable corresponding feature encoding method to determine the feature vector corresponding to the scene behavior data. The model application module is used to train the user profile information and the feature vector using a deep learning algorithm to obtain a trained user intention recognition model, so as to predict the intention of the user to be identified for the target business scenario through the user intention recognition model. The variables include network connection variables and category variables, and the feature representation module includes a first encoding module, a second encoding module, and a feature concatenation module; The first encoding module can be configured to: count the values ​​of network connection variables contained in the scene behavior data to obtain a first vector of the network connection variables; The second encoding module is configured to: encode the category variables contained in the scene behavior data using a one-hot encoding method to obtain the second vector corresponding to the category variables; The feature concatenation module is used to concatenate the first vector and the second vector to obtain the feature vector corresponding to the scene behavior data.

9. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the user intention recognition method according to any one of claims 1 to 6 by executing the executable instructions.

10. A computer-readable storage medium, characterized in that, It stores a computer program, which, when executed by a processor, implements the user intention recognition method according to any one of claims 1 to 6.

11. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium; the processor of the computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the user intention recognition method according to any one of claims 1 to 6.

Images