User quantification method and device based on time sequence, equipment and medium
By extracting and fusion of interactive event time series data, using long and short-term memory networks and hidden Markov models, the shortcomings of traditional user quantization methods in processing complex customer behavior data are solved, and more efficient user quantization model performance is achieved.
Patent Information
- Application Number
- CN202411919416.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-22
AI Technical Summary
Traditional user quantification methods rely solely on simple statistical indicators and intuitive data analysis, and cannot effectively process complex customer behavior data.
By obtaining the time series data of interactive events, preset feature extraction models such as long-term short-term memory network and hidden Markov model are used for feature extraction, and sequence feature information is fused with basic feature information to generate fusion feature data, and finally obtain user quantization results through the trained user quantization model.
It improves the applicability and prediction ability of the user quantitative model when processing complex customer behavior data, and significantly improves the accuracy of the user quantitative model.
Smart Images
Figure CN119862530A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a user quantification method, device, equipment and medium based on time series. Background Art
[0002] In marketing and customer relationship management, in-depth understanding and quantification of user behavior is the key to improving customer service quality and marketing effectiveness. User quantification models are designed to distinguish the characteristics and behavior patterns of different user groups by analyzing customer-related feature data, thereby achieving more accurate customer stratification and personalized services.
[0003] At present, traditional user quantification methods often rely only on simple statistical indicators and intuitive data analysis, which are not suitable for processing complex customer behavior data. Summary of the invention
[0004] The purpose of the embodiments of the present application is to provide a time series-based user quantification method, device, equipment and medium to improve the applicability of the user quantification model when processing complex customer behavior data.
[0005] In a first aspect, an embodiment of the present application provides a user quantification method based on a time series, comprising:
[0006] Obtain the interaction event time series data corresponding to the target user;
[0007] Extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user;
[0008] Fusing the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data;
[0009] A user quantification result corresponding to the target user is obtained based on the fused feature data.
[0010] In an embodiment of the present application, by constructing the interaction event time series data of the target user, extracting the features of the sequence data and fusing them into the basic features of the user, and obtaining the user's quantitative results based on the fused feature data, the applicability of the user quantitative model to process complex customer behavior data can be effectively improved.
[0011] In some possible embodiments, the interaction event time series data includes an event action time series and a dialogue feature time series;
[0012] The extracting features of the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes:
[0013] Based on a preset feature extraction model, feature extraction is performed on the event action time sequence and the conversation feature time sequence respectively to obtain sequence feature information corresponding to the target user.
[0014] In an embodiment of the present application, the performance of the user quantification model in processing complex customer behavior data is further improved by dividing the interaction event time series data into event action time series and conversation feature time series, and performing feature extraction on each of them.
[0015] In some possible embodiments, extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes:
[0016] A preset long short-term memory network is used to extract features from the interaction event time series data, and an output result of the penultimate layer in the long short-term memory network is used as sequence feature information corresponding to the target user.
[0017] In an embodiment of the present application, by utilizing a long short-term memory network to perform feature extraction on time series data, long-term dependencies and patterns in the data can be extracted, thereby further improving the performance of the user quantification model in processing complex customer behavior data.
[0018] In some possible embodiments, extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes:
[0019] A preset hidden Markov model is used to extract features from the interaction event time series data, and a state transition probability result output by the hidden Markov model is used as sequence feature information corresponding to the target user.
[0020] In the embodiment of the present application, by using the hidden Markov model to extract features from time series data, it is possible to extract implicit state information in the data, thereby further improving the performance of the user quantification model in processing complex customer behavior data.
[0021] In some possible embodiments, the sequence feature information is fused with the basic feature information corresponding to the target user to obtain fused feature data, including:
[0022] Determining a feature representation format corresponding to the basic feature information of the target user;
[0023] Convert the sequence feature information into a format according to the feature representation format to obtain feature data to be fused corresponding to the sequence feature information;
[0024] The feature data to be fused is fused with the basic feature information corresponding to the target user to obtain fused feature data.
[0025] In an embodiment of the present application, the features extracted from the time series data are formatted according to the format of the basic features to achieve fusion of the basic features and the time series, thereby further improving the performance of the user quantification model in processing complex customer behavior data.
[0026] In some possible embodiments, acquiring a user quantification result corresponding to the target user based on the fused feature data includes:
[0027] Inputting the fused feature data into a trained user quantification model to obtain a user quantification result corresponding to the target user output by the user quantification model;
[0028] The user quantification model is obtained by training based on pre-collected fusion feature sample data and its corresponding user quantification annotation results.
[0029] In an embodiment of the present application, the user quantification model is trained to output corresponding user quantification results according to the fusion features, thereby further improving the performance of the user quantification model in processing complex customer behavior data.
[0030] In some possible embodiments, the training method of the user quantification model includes: obtaining time series data of historical interaction events corresponding to a number of historical users;
[0031] Performing feature extraction on the historical interaction event time series data based on the feature extraction model to obtain historical sequence feature information corresponding to the historical user;
[0032] Fusing the historical sequence feature information with the basic feature information corresponding to the historical user to obtain corresponding fused feature sample data;
[0033] The preset classification model is trained based on the fused feature sample data corresponding to each historical user and the corresponding user quantitative annotation results to obtain the trained user quantitative model.
[0034] In an embodiment of the present application, features are extracted from time series data of historical users, and the classification model is trained after the extracted features are fused with the corresponding basic features to obtain a trained user quantification model, thereby further improving the performance of the user quantification model in processing complex customer behavior data.
[0035] In a second aspect, an embodiment of the present application provides a user quantification device based on a time series, comprising:
[0036] A data acquisition module is used to acquire the time series data of interaction events corresponding to the target user;
[0037] A feature extraction module, used to extract features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user;
[0038] A feature fusion module, used to fuse the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data;
[0039] A user quantification module is used to obtain a user quantification result corresponding to the target user based on the fused feature data.
[0040] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method described in any embodiment of the first aspect when executing the program.
[0041] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented.
[0042] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a computer program, wherein when the computer program is executed by a processor, the method described in any embodiment of the first aspect can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.
[0044] Figure 1 A schematic diagram of a flow chart of a user quantification method based on time series provided in an embodiment of the present application;
[0045] Figure 2 A schematic diagram of the structure of a user quantification device based on time series provided in an embodiment of the present application;
[0046] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0047] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.
[0048] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.
[0049] It should be noted that traditional user quantification solutions often rely only on simple statistical indicators and intuitive data analysis, lacking the ability to mine deeper features of the data, and are therefore not suitable for processing complex customer behavior data.
[0050] In view of the problems existing in the above-mentioned prior art, an embodiment of the present application provides a user quantification method based on time series, which extracts features from the time series data of the user's interaction events and integrates them into the user's basic features before performing user quantification, thereby being able to mine deeper features of the data and effectively improve the applicability of the user quantification model in processing complex customer behavior data.
[0051] like Figure 1 As shown, the embodiment of the present application provides a user quantification method based on time series, which may include the steps of:
[0052] S1. Obtain interaction event time series data corresponding to the target user.
[0053] Specifically, during the telephone sales process, various data generated during the interaction between the AI robot and the user, such as interaction events and dialogue content nodes and their time information, can be recorded in real time to generate interaction event time series data corresponding to each user. When it is necessary to conduct user quantitative analysis on the target user, the interaction event time series data corresponding to the target user is obtained.
[0054] S2. Extract features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user.
[0055] Specifically, model training can be performed based on historically collected interaction event time series data to obtain a trained feature extraction model, which is used to extract features from current interaction event time series data to obtain sequence feature information corresponding to the current target user.
[0056] S3. Fusing the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data.
[0057] Specifically, after obtaining the sequence feature information extracted based on the interaction event time series data of the target user, the sequence feature information is fused with the pre-acquired basic feature information of the target user to obtain fused feature data.
[0058] It is understandable that the basic characteristic information of a user refers to the characteristic information quantified by the traditional method. For example, the basic characteristic information may be objective characteristics such as age, gender, city level, or characteristic information such as call duration and communication rounds generated during telephone communication. The specific type of basic characteristic information can be set according to demand.
[0059] S4. Obtain a user quantification result corresponding to the target user based on the fused feature data.
[0060] Finally, the user quantification result of the target user can be obtained based on the fused feature data. Exemplarily, a model training method can be used to train a model through pre-labeled data to learn the relationship between the fused feature data and the user quantification result.
[0061] Based on this, compared with the traditional solution of user quantification that only relies on simple statistical indicators and intuitive data, the embodiment of the present application effectively improves the predictive ability of the user quantification model by further integrating the sequence features extracted based on time series data, thereby effectively improving the applicability of the user quantification model to process complex customer behavior data.
[0062] In some possible embodiments, the interaction event time series data includes an event action time series and a conversation feature time series;
[0063] Step S2, extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user, may include:
[0064] S201 . Based on a preset feature extraction model, feature extraction is performed on the event action time series and the conversation feature time series respectively to obtain sequence feature information corresponding to the target user.
[0065] It should be noted that the user's interaction event time series data may include one or more sequences. Exemplarily, in the process of telephone sales for a certain user, the interaction event time series data formed may include two sequences: 1. Event action time series: recording the interaction events between the AI robot and the user and their time points, such as IVR_26 (the event action of the AI robot at the 26th second) and USER_28 (the event action of the user at the 28th second); 2. Dialogue feature time series: recording the key events and time points during the call, such as connected_14 (the call was connected at the 14th second) and customer hung up_66 (the customer hung up the phone at the 66th second).
[0066] It can be understood that when the interaction event time series data includes more than one type of sequence data, considering that different sequence types have different characteristic features, model training can be performed separately for different sequence data in the model training stage, and then the feature extraction model corresponding to each sequence type is obtained in the feature recognition stage to perform feature extraction on the corresponding sequence data.
[0067] Based on this, the accuracy of feature extraction of interaction event time series data is further improved by dividing the interaction event time series data into event action time series and dialogue feature time series, and using corresponding models for feature extraction respectively.
[0068] In some possible embodiments, step S2, extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user, may include:
[0069] S211. Use a preset long short-term memory network to extract features from the interaction event time series data, and use the output result of the second-to-last layer in the long short-term memory network as the sequence feature information corresponding to the target user.
[0070] Optionally, for the interactive event time series data, a pre-trained long short-term memory network (LSTM) can be used to extract its features. The output result of the second-to-last layer in the long short-term memory network is used as the sequence feature information of the interactive event time series data; it can be understood that the final prediction result output by the LSTM network model in the last layer and the output result of the second-to-last layer represent the high-level features extracted by LSTM when processing sequence data. These features can be regarded as a compressed representation of the input time series. By extracting this feature information, important patterns, trends and dependencies in the sequence can be captured.
[0071] Exemplarily, the process of extracting features from the LSTM network model may include the following process:
[0072] 1. Data Preparation
[0073] 1. Coding events:
[0074] Encode the event type. For example, use one-hot encoding to create a vector representation for each event.
[0075] For example, the "connection" event in the time series is coded as [1,0,0,...]; the "verification" event is coded as [0,1,0,...]... and so on, the time series can be converted into a numerical series.
[0076] 2. Sequence construction:
[0077] The above encoded sequence is converted into a fixed-length time window and input into the LSTM. For example, a sequence of 10-15 events can be selected as input; those with insufficient length need to be padded with 0.
[0078] 2. Feature Extraction
[0079] 1. Build LSTM model:
[0080] Define the LSTM model structure, which usually includes an input layer, several LSTM layers, a fully connected layer, and an output layer. The format of the input layer can be defined as [batch size, time step, feature dimension].
[0081] 2. Model training:
[0082] The LSTM model is trained using pre-collected training data (for example, the input is an event sequence and the target is whether the user purchases insurance).
[0083] After the model training is completed, the output of the penultimate layer can be extracted as the feature information of the sequence.
[0084] For example, the features extracted by LSTM may include:
[0085] 1. Behavior pattern:
[0086] Identify the user's behavioral patterns during a phone call, such as whether the "authentication" event occurs frequently after the call is connected, the user's reaction to "opening WeChat", etc.
[0087] 2. User attitude:
[0088] Extracting changes in user attitudes, such as the number of occurrences of “clearly negative” events and their timing, may affect the final purchase decision.
[0089] 3. Time dependence:
[0090] For example, the time point at which the "user hangs up" event occurs and the sequence of previous events at this time can reflect the user's interest in the sale.
[0091] In some possible embodiments, step S2, extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user, may include:
[0092] S221. Use a preset hidden Markov model to extract features from the interaction event time series data, and use the state transition probability result output by the hidden Markov model as the sequence feature information corresponding to the target user.
[0093] It should be noted that in addition to using the LSTM network model to extract features from the interactive event time series data, the hidden Markov model can also be used to extract features from the interactive event time series data. It is understandable that in specific applications, one of the LSTM network model and the hidden Markov model or both models can be used to extract features from the interactive event time series data.
[0094] It can be understood that the Hidden Markov Model (HMM) is a statistical model used to describe Markov processes with hidden unknown parameters. Among them, the implicit state refers to the state that we cannot directly observe, such as the user's purchase intention state (such as "hesitation", "rejection", "considering", etc.); the observed event refers to the event that we can observe, such as the user's behavior on the phone (such as "connected", "verified", "hang up", etc.); the state transition probability refers to the probability of transitioning from one implicit state to another implicit state.
[0095] The following are specific examples to illustrate the input and output process of the HMM model:
[0096] For example, assume that the currently defined implicit states and observed events are as follows:
[0097] 1. Implicit state (implicit user intent):
[0098] s1: high purchase intention (positive);
[0099] s2: hesitation;
[0100] s3: Reject.
[0101] 2. Observed events (actual events):
[0102] e1: connected;
[0103] e2: core body;
[0104] e3: User attitude - clear negative;
[0105] e4: The customer hangs up.
[0106] Assume that the interaction event sequence of the target user currently obtained is as follows: ["connect", "identity verification", "user attitude - clear denial", "customer hangs up"].
[0107] Input the above sequence into the HMM model, and the model will calculate the hidden state corresponding to each observation. For example, the first observation event "connection" may correspond to state s1 (high purchase intention); the second observation event "verification" still corresponds to s1; the third observation event "user attitude-clear denial" may go to state s3 (rejection); the last observation event "customer hangs up" also corresponds to state s3 (rejection). Therefore, the possible hidden state sequence is: ["s1","s1","s3","s3"].
[0108] Assuming that the current focus is on the probability of going from "undealt" (such as state s3) to "dealt", the model will calculate that the probability of going from state s3 (rejection) to state deal is P(deal|rejection) = 0.1 (example). This probability value is the characteristic information output by the HMM model for the event sequence.
[0109] It can be understood that through the HMM model, the event sequence of the user in the conversation can be converted into an implicit purchase intention state, and the probability of going from an untransacted state to a transacted state can be obtained. These probability values can be used as valuable features to further predict whether the user will purchase insurance.
[0110] In some possible embodiments, step S3, fusing the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data, may include:
[0111] S301, determining a feature representation format corresponding to basic feature information of a target user;
[0112] S302, converting the format of the sequence feature information according to the feature representation format to obtain feature data to be fused corresponding to the sequence feature information;
[0113] S303: Fusing the feature data to be fused with the basic feature information corresponding to the target user to obtain fused feature data.
[0114] Specifically, in the process of fusing the sequence feature information with the basic feature information, the feature representation format of the basic feature information (such as string type and length, vector conversion method, etc.) can be first determined, and then the sequence feature information extracted in the above steps is format converted according to the same feature representation format to obtain feature data to be fused that has the same representation form as the basic feature information; finally, the feature data to be fused is fused with the basic feature information to obtain fused feature data.
[0115] Exemplarily, feature fusion can be directly superimposing the feature data to be fused with the basic feature information. For example, assuming that the basic feature information of the user contains 100 features, represented as (f1, f2...f100), the feature data to be fused is represented as (T11, T12, T21, T22) through feature extraction and format conversion of the user's interactive event time series data, where T11 and T21 are sequence features extracted and converted by the LSTM network model for the event action time series and the dialogue feature time series, respectively, and T12 and T22 are sequence features extracted and converted by the HMM model for the event action time series and the dialogue feature time series, respectively; then the fused feature data obtained by fusing the basic feature information and the feature data to be fused can be represented as f1, f2...f100)+(T11, T12, T21, T22).
[0116] Based on this, compared with the traditional solution that only relies on basic feature information for user quantification, the embodiment of the present application utilizes the deep information in the time series data by integrating the sequence feature information, thereby significantly improving the prediction ability and accuracy of the user quantification model.
[0117] In some possible embodiments, step S4, obtaining a user quantification result corresponding to the target user based on the fused feature data, may include:
[0118] S401, inputting the fused feature data into a trained user quantification model to obtain a user quantification result corresponding to a target user output by the user quantification model;
[0119] Among them, the user quantification model is trained based on the pre-collected fusion feature sample data and its corresponding user quantification annotation results.
[0120] Furthermore, the training method of the user quantization model may include:
[0121] S411, obtaining time series data of historical interaction events corresponding to a number of historical users;
[0122] S412, extracting features from the historical interaction event time series data based on the feature extraction model to obtain historical sequence feature information corresponding to the historical user;
[0123] S413, fusing the historical sequence feature information with the basic feature information corresponding to the historical user to obtain corresponding fused feature sample data;
[0124] S414: Train the preset classification model based on the fused feature sample data corresponding to each historical user and the corresponding user quantitative annotation results to obtain a trained user quantitative model.
[0125] Exemplarily, the user quantification result may be a result of classifying users according to the fused feature data, for example, into high-quality users and non-high-quality users, etc. The process of obtaining the user quantification result may be achieved by training a user quantification model.
[0126] Specifically, based on steps S1 to S3, the historically collected user data can be fused with feature data for extraction (to obtain fused feature sample data), and then the user data are labeled separately according to actual conditions to obtain corresponding user quantitative labeling results; finally, the preset classification model is trained based on the fused feature sample data and the corresponding user quantitative labeling results to obtain a trained user quantitative model.
[0127] Please refer to Figure 2 , Figure 2 FIG. 1 shows a block diagram of a user quantification device based on a time series provided in some embodiments of the present application. It should be understood that the user quantification device based on a time series is similar to the above-mentioned Figure 1 Corresponding to the method embodiment, each step involved in the above method embodiment can be executed. The specific functions of the time series-based user quantization device can refer to the above description. To avoid repetition, the detailed description is appropriately omitted here.
[0128] Figure 2 The user quantification device based on time series includes at least one software function module that can be stored in a memory in the form of software or firmware or solidified in the user quantification device based on time series, and the user quantification device based on time series includes:
[0129] The data acquisition module 210 is used to acquire the interaction event time series data corresponding to the target user;
[0130] A feature extraction module 220 is used to extract features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user;
[0131] A feature fusion module 230 is used to fuse the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data;
[0132] The user quantification module 240 is used to obtain a user quantification result corresponding to the target user based on the fused feature data.
[0133] It can be understood that the above-mentioned device embodiment corresponds to the method embodiment of the present invention. The user quantification device based on time series provided by the embodiment of the present invention can implement the user quantification method based on time series provided by any method embodiment of the present invention.
[0134] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method, and will not be described in detail here.
[0135] like Figure 3 As shown, some embodiments of the present application provide an electronic device 300, which includes: a memory 310, a processor 320, and a computer program stored in the memory 310 and executable on the processor 320, wherein the processor 320 reads the program from the memory 310 through a bus 330 and executes the program to implement a method of any embodiment of the user quantization method based on time series as described above.
[0136] Processor 320 can process digital signals and can include various computing structures, such as complex instruction set computer structure, reduced instruction set computer structure, or a structure that implements a combination of multiple instruction sets. In some examples, processor 320 can be a microprocessor.
[0137] The memory 310 may be used to store instructions executed by the processor 320 or data related to the execution of instructions. These instructions and / or data may include codes for implementing some or all functions of one or more modules described in the embodiments of the present application. The processor 320 of the disclosed embodiment may be used to execute instructions in the memory 310 to implement the method shown above. The memory 310 includes a dynamic random access memory, a static random access memory, a flash memory, an optical memory, or other memory known to those skilled in the art.
[0138] Some embodiments of the present application further provide a computer-readable storage medium having a computer program stored thereon. The computer program is executed by a processor to execute the method described in the method embodiment.
[0139] Some embodiments of the present application further provide a computer program product, which, when executed on a computer, enables the computer to execute the method described in the method embodiment.
[0140] It should be noted that each embodiment in this specification is described in a progressive manner, and each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.
[0141] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present application. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.
[0142] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0143] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.
[0144] The above description is only an embodiment of the present application and is not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.
[0145] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0146] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.
Claims
1. A user quantification method based on time series, characterized in that: include: Obtain the interaction event time series data corresponding to the target user; Extracting features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user; Fusing the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data; A user quantification result corresponding to the target user is obtained based on the fused feature data.
2. The user quantification method based on time series according to claim 1, characterized in that: The interaction event time series data includes event action time series and dialogue feature time series; The extracting features of the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes: Based on a preset feature extraction model, feature extraction is performed on the event action time sequence and the conversation feature time sequence respectively to obtain sequence feature information corresponding to the target user.
3. The user quantification method based on time series according to claim 1, characterized in that: The extracting features of the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes: A preset long short-term memory network is used to extract features from the interaction event time series data, and an output result of the penultimate layer in the long short-term memory network is used as sequence feature information corresponding to the target user.
4. The user quantification method based on time series according to claim 1, characterized in that: The extracting features of the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user includes: A preset hidden Markov model is used to extract features from the interaction event time series data, and a state transition probability result output by the hidden Markov model is used as sequence feature information corresponding to the target user.
5. The user quantification method based on time series according to claim 1, characterized in that: The sequence feature information is fused with the basic feature information corresponding to the target user to obtain fused feature data, including: Determining a feature representation format corresponding to the basic feature information of the target user; Convert the sequence feature information into a format according to the feature representation format to obtain feature data to be fused corresponding to the sequence feature information; The feature data to be fused is fused with the basic feature information corresponding to the target user to obtain fused feature data.
6. The user quantification method based on time series according to claim 1, characterized in that: The acquiring a user quantification result corresponding to the target user based on the fused feature data includes: Inputting the fused feature data into a trained user quantification model to obtain a user quantification result corresponding to the target user output by the user quantification model; The user quantification model is obtained by training based on pre-collected fusion feature sample data and its corresponding user quantification annotation results.
7. The user quantification method based on time series according to claim 6, characterized in that: The training method of the user quantification model includes: obtaining time series data of historical interaction events corresponding to a number of historical users; Performing feature extraction on the historical interaction event time series data based on the feature extraction model to obtain historical sequence feature information corresponding to the historical user; Performing feature fusion on the historical sequence feature information and the basic feature information corresponding to the historical user to obtain corresponding fused feature sample data; The preset classification model is trained based on the fused feature sample data corresponding to each historical user and the corresponding user quantitative annotation results to obtain the trained user quantitative model.
8. A user quantification device based on time series, characterized in that: include: A data acquisition module is used to acquire the time series data of interaction events corresponding to the target user; A feature extraction module, used to extract features from the interaction event time series data based on a preset feature extraction model to obtain sequence feature information corresponding to the target user; A feature fusion module, used to fuse the sequence feature information with the basic feature information corresponding to the target user to obtain fused feature data; A user quantification module is used to obtain a user quantification result corresponding to the target user based on the fused feature data.
9. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the user quantification method based on time series described in any one of claims 1 to 7 can be implemented when the processor executes the program.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the time series-based user quantification method according to any one of claims 1 to 7 is executed.