CN112364663B - User characteristic identification method, device, equipment and storage medium - Google Patents
User characteristic identification method, device, equipment and storage medium Download PDFInfo
- Publication number
- CN112364663B CN112364663B CN202011281605.4A CN202011281605A CN112364663B CN 112364663 B CN112364663 B CN 112364663B CN 202011281605 A CN202011281605 A CN 202011281605A CN 112364663 B CN112364663 B CN 112364663B
- Authority
- CN
- China
- Prior art keywords
- text
- feature
- training data
- content
- user
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/23—Updating
- G06F16/2308—Concurrency control
- G06F16/2315—Optimistic concurrency control
- G06F16/2322—Optimistic concurrency control using timestamps
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
Abstract
The embodiment of the application provides a user characteristic identification method, device and equipment and a storage medium. According to the method, the plurality of first text information of the user is obtained, the first text information comprises first text content and time stamps corresponding to the first text content, at least one feature of the user is determined according to the plurality of first text information, at least one feature of the user is output, and the at least one feature of the user is determined by combining the plurality of first text content related to the user and the time stamps corresponding to each first text content, so that the feature based on a time dimension can be obtained, and the obtained feature has stronger representation capability.
Description
Technical Field
The embodiment of the application relates to the technical field of data processing, and more particularly relates to a method, a device, equipment and a storage medium for identifying user characteristics.
Background
Features of the user may be identified based on the user information in order to provide targeted services to the user. The user information may be a time series text, for example, the user information is obtained by collecting user feedback, user investigation, user behavior, user language, and the like.
At present, collected user information is spliced according to a sequence, and then spliced texts are analyzed to obtain user characteristics.
Disclosure of Invention
The embodiment of the application provides a user characteristic identification method, device and equipment and a storage medium.
In a first aspect, a method for identifying a user feature is provided, including:
acquiring a plurality of first text messages of a user, wherein the first text messages comprise first text contents and timestamps corresponding to the first text contents;
determining at least one feature of the user based on the plurality of first text messages;
at least one characteristic of the user is output.
In a specific implementation, determining at least one feature of the user according to the plurality of first text information includes:
mapping, for each first text message, the first text message into a sequence of text vectors, the sequence of text vectors being used to characterize the first text content and the time information corresponding to the first text content in a vector space;
and determining at least one characteristic of the user according to a plurality of text vector sequences corresponding to the plurality of first text information.
In a specific implementation, mapping the first text information into a sequence of text vectors for each first text information includes:
mapping the first text content of the first text information into a sequence of content vectors, the sequence of content vectors being used to characterize the first text content in a vector space;
mapping the time stamp corresponding to the first text content into a time vector sequence according to a preset time dimension, wherein the time dimension comprises at least one of year, month, day, week, hour or minute, and the time vector sequence is used for representing the time information of the first text content in the time dimension in a vector space;
and fusing the content vector sequence and the time vector sequence to obtain a text vector sequence.
Optionally, determining at least one feature of the user according to the plurality of first text information includes:
and inputting a plurality of first text messages into the pre-trained feature model to obtain at least one feature of the user.
Optionally, before inputting the plurality of first text information into the feature model to obtain the at least one feature of the user, the method further comprises:
acquiring a first training data set, wherein the first training data set comprises a plurality of first training data, the first training data comprises a plurality of second text information and at least one characteristic label, and the second text information comprises second text content and a timestamp corresponding to the second text content;
and training the feature model through the first training data set.
Optionally, before training the feature model by the first training data set, the method further comprises:
acquiring a second training data set, wherein the second training data set comprises a plurality of second training data, and the second training data comprises third text content;
training a text model through a second training data set, wherein the text model comprises a semantic generation module and a text prediction module, and a full connection layer in the text prediction module receives a first semantic vector and a plurality of second semantic vectors which are output by the semantic generation module, and the first semantic vector comprises semantic information corresponding to the plurality of second semantic vectors;
after the loss function of the text model is converged to a preset value, a feature model is generated, wherein the feature model comprises a semantic generation module and a feature prediction module, and a full connection layer in the feature prediction module receives a first semantic vector output by the semantic generation module.
Further, training the text model with the second training data set, comprising:
performing mask processing on at least one element in the third text content by a text model aiming at each second training data in the second training data set to obtain mask processed second training data, wherein the mask processed second training data comprises mask processed third text content and a content tag, and the content tag is at least one element replaced by the mask;
training the text model through the second training data after the plurality of mask processing.
In a second aspect, there is provided an apparatus for identifying a user feature, comprising:
an acquisition unit configured to acquire a plurality of first text information of a user, the first text information including first text content and a timestamp corresponding to the first text content;
a processing unit for determining at least one feature of the user based on the plurality of first text information;
and an output unit for outputting at least one characteristic of the user.
In a third aspect, there is provided an electronic device comprising: a processor and a memory for storing a computer program, the processor being for invoking and running the computer program stored in the memory for performing the method as in the first aspect or implementations thereof.
In a fourth aspect, a computer-readable storage medium is provided for storing a computer program for causing a computer to perform the method as in the first aspect or in various implementations thereof.
By means of the technical scheme of the first aspect, at least one feature of the user can be determined by combining the acquired plurality of first text contents related to the user and the time stamp corresponding to each first text content, the feature based on the time dimension can be obtained, and the obtained feature has stronger representation capability.
Drawings
Fig. 1 is a flowchart of a method 100 for identifying user features according to an embodiment of the present application;
fig. 2 is a schematic diagram of a first text information processing process 200 according to an embodiment of the present application;
fig. 3 is a schematic diagram of a first text information processing process 300 according to an embodiment of the present application;
FIG. 4 is a schematic structural diagram of a feature model 400 according to an embodiment of the present disclosure;
fig. 5 is a flowchart of a method 500 for identifying user features according to an embodiment of the present application;
FIG. 6 is a schematic diagram of a word embedding process 600 for a timestamp according to an embodiment of the present application;
FIG. 7 is a schematic diagram of a feature model training process 700 according to an embodiment of the present application;
FIG. 8 is a schematic diagram of a feature model training process 800 according to an embodiment of the present disclosure;
fig. 9 is a schematic structural diagram of a text model 900 according to an embodiment of the present application;
fig. 10 is a schematic hardware structure of an electronic device 1000 according to an embodiment of the present application.
Detailed Description
The following description of the technical solutions in the embodiments of the present application will be made with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art without undue burden for the embodiments herein, are intended to be within the scope of the present application.
At present, a time sequence text of a neural network model is input, and only a sequence is present without specific time information, so that the identified user features do not have the features of the user in the time dimension, and the user features have a relatively poor representation capability for the user, for example, the features of the user at the same time every day or the features of every monday cannot be identified.
In order to solve the technical problems, the invention concept of the application is as follows: and carrying out vector characterization and vector fusion on the collected text information related to the user, including the text content related to the user and the timestamp corresponding to each text content, so as to output at least one characteristic of the user, wherein the at least one characteristic comprises characteristics of the user in one or more time dimensions. Illustratively, the text content associated with the user includes a utterance made by the user at the social platform, search content entered via a search engine, transaction records generated by making payments over a network, and so forth.
The technical scheme of the embodiment of the application can be suitable for each collected text content which is in a period of time or is related to a certain amount of users, namely, the user characteristics are dynamically identified, and at least one current characteristic of the users is obtained.
It should be appreciated that in embodiments of the present application, the terminal device may be a cellular telephone, a cordless telephone, a session initiation protocol (Session Initiation Protocol, SIP) phone, a wireless local loop (Wireless Local Loop, WLL) station, a personal digital assistant (Personal Digital Assistant, PDA) device, a handheld device having wireless communication capabilities, a computing device or other processing device connected to a wireless modem, an in-vehicle device, a wearable device, etc., without limitation.
The technical scheme of the embodiment of the present application will be described in detail below:
in order to identify the characteristics of the user in the time dimension, the embodiment of the application combines the time stamp corresponding to the text content related to the user to identify the characteristics of the user through vector characterization and vector fusion.
Fig. 1 is a flowchart of a method 100 for identifying user features according to an embodiment of the present application. The main execution body of the method may be part or all of the terminal device, where part of the terminal device may be a processor of the terminal device, and the main execution body of the method may also be the terminal device and the server, that is, in fig. 1, a part of steps are executed by the terminal device, and another part of steps are executed by the server, which is not limited in this application. As shown in fig. 1, the method includes:
s101: a plurality of first text information of a user is acquired.
Wherein the first text information includes first text content and a timestamp corresponding to the first text content. By way of example, the first text content may be a utterance made by a user at a social platform, search content entered via a search engine, transaction records resulting from making payments over a network, and so forth; the time stamp corresponding to each first text content is the publishing time, the output time or the recording time of the corresponding first text content, and the like.
For example, the method may periodically collect a plurality of first text information of the user, or collect a plurality of first text information within a preset time period, or collect a preset number of first text information, which is not limited in this application.
S102: at least one feature of the user is determined based on the plurality of first text information.
In this step, the user's characteristics are determined not only from the first text content, but also in combination with the first text content and the corresponding time stamp, at least one characteristic of the user is determined. It should be appreciated that some or all of the at least one feature is a feature of a time dimension, such as a user retrieving a workout video via a search engine, and both times are morning, then determining that the at least one feature of the user should include "workout in morning".
For example, at least one feature of the user may be output through vector characterization and vector fusion. Fig. 2 is a schematic diagram of a first text information processing process 200 according to an embodiment of the present application; fig. 3 is a schematic diagram of a first text information processing process 300 according to an embodiment of the present application.
Referring to fig. 2, each first text message is mapped into a text vector sequence, each text vector sequence is used for representing the first text content and corresponding time information of the first text message in a vector space, and at least one feature of the user is determined according to the obtained text vector sequence corresponding to each first text message.
Illustratively, each text vector sequence includes a first text vector and a plurality of second text vectors. Each second text vector is used for representing the time-space information, content or type and the like of any element in the first text content, and the time-space information is used for representing the time sequence and/or space coordinates of the element in the first text content; the first text vector contains all information of the plurality of second text vectors. Further, the at least one feature of the user is determined from the plurality of text vector sequences, and in particular, the at least one feature of the user may be determined from the first text vector in each text vector sequence. Alternatively, the first text vector may be the first vector in the sequence of text vectors.
As shown in fig. 3, mapping each first text message into a text vector sequence specifically includes: mapping first text content of the first text information into a content vector sequence, mapping a time stamp corresponding to the first text content into a time vector sequence according to a preset time dimension; and carrying out vector fusion on the content vector sequence and the time vector sequence to obtain a text vector sequence.
It will be appreciated that the content vector sequence is used to characterize the first text content in a vector space, and the temporal vector sequence is used to characterize the temporal information of the first text content in a temporal dimension in a vector space. The time dimension includes, but is not limited to, at least one of a year, month, day, week, hour, or minute.
Alternatively, vector fusion of the content vector sequence and the time vector sequence may specifically be summation of two vectors.
S103: at least one characteristic of the user is output.
In the embodiment of the application, at least one feature of the user can be determined by combining the acquired plurality of first text contents related to the user and the time stamp corresponding to each first text content, so that the feature based on the time dimension can be obtained, and the obtained feature has stronger representation capability.
In the embodiments of the present application: the input first text information is subjected to vector characterization and vector fusion through the feature model so as to output at least one feature of the user, wherein the output feature of the user at least comprises user features in one time dimension.
The feature model comprises a semantic generation module and a feature prediction module, wherein the semantic generation module is used for carrying out semantic recognition on a plurality of input first text information, and the feature prediction module predicts and obtains the features of the user according to a semantic vector sequence output by the semantic generation module.
Fig. 4 is a schematic structural diagram of a feature model 400 according to an embodiment of the present application. As shown in fig. 4, the semantic generation module includes 4 data conversion transform structures and an Embedding layer (Embedding). Illustratively, each transducer structure includes: self-attention mechanism layers, such as Masked multi Self Attention layers; a normalization Layer (Layer Norm) connected with the self-attention mechanism Layer, wherein the normalization Layer is used for regularizing the vector input by the self-attention mechanism Layer; a Feed Forward Layer (Feed Forward) coupled to the Layer Norm, the Feed Forward comprising two fully coupled layers; another Layer Norm connected to Feed Forward.
The embedding layer is used for word embedding of the input text information, and for example, when the input text information comprises text content and a time stamp, word embedding can be respectively carried out on the text content and the time stamp.
Fig. 5 is a schematic flow chart of a method 500 for identifying user features according to an embodiment of the present application. As shown in fig. 5, a plurality of first text messages of a user are input into a feature model, for each first text message, an input module of the feature model obtains corresponding time information according to a preset time dimension according to a timestamp in the first text message, then the first text content and the time information in the first text message are input into an embedding layer of a semantic generation module, the embedding layer respectively performs word embedding on the first text content and the time information to obtain a content vector sequence and a time vector sequence, the content vector sequence and the time vector sequence are subjected to vector fusion to obtain a text vector sequence, semantic recognition is performed on the text vector sequence through 4 Transformer structures, and at least one feature of the user is obtained through a feature prediction module according to a semantic recognition result.
Fig. 6 is a schematic diagram of a word embedding process 600 of a timestamp according to an embodiment of the present application. As shown in fig. 6, for each timestamp in the first text information, according to preset time dimensions, such as week, hour and date, time information corresponding to each time dimension is obtained, word embedding is performed on the time information of each time dimension to obtain a time vector sequence of each time dimension, and vector fusion is performed on the time vector sequence of each dimension to obtain a time vector sequence corresponding to the first text information.
The following will focus on how to perform vector characterization and vector fusion on the input multiple first text information through the feature model to obtain a training process of the feature model.
The training of the feature model can be classified into unsupervised training and supervised training.
The embodiment of the present application is illustrated by taking supervised training as an example, and fig. 7 is a schematic diagram of a feature model training process 700 provided in the embodiment of the present application.
First, a first training data set for training a feature model is obtained, the first training data set comprising a plurality of second text information and at least one feature tag, wherein the second text information comprises second text content and a corresponding timestamp.
And combining the text vector sequences, carrying out semantic recognition on the text vector sequences to obtain semantic vectors, determining at least one feature of a user according to the semantic vectors, calculating a loss value of the feature model according to the obtained feature and at least one feature tag in the corresponding first training data through a preset loss function, ending the training process after the loss value converges to the preset value, and obtaining the feature model with higher accuracy.
Based on the embodiment, the feature model in the embodiment of the application can be converted from a language model, so that the feature model has higher semantic recognition capability.
Fig. 8 is a schematic diagram of a feature model training process 800 according to an embodiment of the present application. As shown in fig. 8, before training the feature model by the first training data set, training the text model, it should be understood that the text model is any language model constructed in advance, and in conjunction with fig. 9, the text model includes a semantic generation module and a text prediction module, where the semantic generation module includes 4 data conversion fransformer structures and an Embedding layer (Embedding), and each of the fransformer structures is terminated, that is, the same fransformer structure repeatedly performs four corresponding processes. Illustratively, each transducer structure includes: self-attention mechanism layers, such as Masked multi Self Attention layers; a normalization Layer (Layer Norm) connected with the self-attention mechanism Layer, wherein the normalization Layer is used for regularizing the vector input by the self-attention mechanism Layer; a Feed Forward Layer (Feed Forward) coupled to the Layer Norm, the Feed Forward comprising two fully coupled layers; another Layer Norm connected to Feed Forward. The text prediction module includes a full connection layer and a loss function.
The text model also includes an input module that may be integrated with the semantic generation module.
The training of the text model to obtain the feature model includes: and acquiring a second training data set, training a text model through the second training data set, and generating a feature model after the loss value of the loss function of the text model is converged to a preset value. Wherein the second training data set comprises a plurality of second training data, the second training data comprising third text content.
It should be understood that the second text content in the first training data set may be the same or different from the third text content in the second training data set, which is not limited in this application.
Referring to fig. 8, for each second training data in the second training data set, the text model performs masking processing on the third text content in each second data to obtain third text content after masking processing, and a content tag corresponding to each third text content, where the content tag is at least one element replaced by a mask in the process of masking processing the third text content.
After masking the third text content by the input module, word embedding is carried out on the masked third text content by the embedding layer, semantic recognition is carried out to obtain a semantic vector, the semantic vector comprises a first semantic vector and a plurality of second semantic vectors, the first semantic vector comprises all semantic information represented by the plurality of second semantic vectors, the text prediction module receives the first semantic vector and the plurality of second semantic vectors output by the semantic generation module by the full connection layer, predicts elements covered by the mask in the third text content according to the first semantic vector and the plurality of second semantic vectors, calculates a loss function according to the elements covered by the mask and content labels obtained by prediction, and ends the training process when the loss value of the loss function converges to a preset value to obtain a trained text model. Further, the text prediction module in the trained text model is replaced by the feature prediction module, so that a feature model is obtained, and the full-connection layer of the feature prediction module receives the first semantic vector output by the semantic generation module, namely, the feature prediction module of the feature model predicts the features of the user only according to the first semantic vector in the semantic vector sequence.
In the process of training the text model to obtain the feature model, the third text content input to the text model is "today's weather is good", the text model carries out mask processing on the third text content to obtain "today's [ M ] weather is good" and "day" of content labels of the third text content after the mask processing, the text model predicts the obtained characters replaced by the mask after word embedding and 4 transform structures, and then the accuracy of the prediction result is determined according to the content labels.
In the embodiment of the application, the language model is used as a basis, the feature model is obtained through training the text model, so that the feature model has the capacity of semantic recognition, and the accuracy of feature prediction is improved through training the feature model, so that the final predicted features of the user are more accurate and reliable.
The method embodiments of the present application are described in detail above in connection with fig. 1 to 8, and the apparatus embodiments of the present application are described in detail below in connection with fig. 9, it being understood that the apparatus embodiments and the method embodiments correspond to each other, and similar descriptions may refer to the method embodiments.
Fig. 9 shows a schematic block diagram of an identification device 900 of a user feature according to an embodiment of the present application. As shown in fig. 9, the apparatus 900 includes:
an acquiring unit 910, configured to acquire a plurality of first text information of a user, where the first text information includes a first text content and a timestamp corresponding to the first text content;
a processing unit 920, configured to determine at least one feature of the user according to the plurality of first text information;
an output unit 930, configured to output at least one feature of the user.
The user feature recognition device 900 in the embodiment of the present application includes an acquisition unit 910, a processing unit 920 and an output unit 930, where at least one feature of a user can be determined by combining a plurality of acquired first text contents related to the user and a timestamp corresponding to each first text content, so that a feature based on a time dimension can be obtained, and the obtained feature has a stronger representation capability.
Optionally, the processing unit 920 is specifically configured to:
mapping, for each first text message, the first text message into a sequence of text vectors for characterizing the first text content and time information corresponding to the first text content in a vector space;
and determining at least one characteristic of the user according to a plurality of text vector sequences corresponding to the plurality of first text information.
Optionally, the processing unit 920 is specifically configured to:
mapping first text content of the first text information into a sequence of content vectors, the sequence of content vectors being used to characterize the first text content in a vector space;
mapping the timestamp corresponding to the first text content into a time vector sequence according to a preset time dimension, wherein the time dimension comprises at least one of year, month, day, week, hour or minute, and the time vector sequence is used for representing the time information of the first text content in the time dimension in a vector space;
and fusing the content vector sequence and the time vector sequence to obtain the text vector sequence.
Optionally, the processing unit 920 is specifically configured to: and inputting the plurality of first text messages into a pre-trained feature model to obtain at least one feature of the user.
Optionally, the obtaining unit 910 is further configured to: acquiring a first training data set, wherein the first training data set comprises a plurality of first training data, the first training data comprises a plurality of second text information and at least one characteristic label, and the second text information comprises second text content and a timestamp corresponding to the second text content; the processing unit 920 is further configured to train the feature model with the first training data set.
Optionally, the obtaining unit 910 is further configured to obtain a second training data set, where the second training data set includes a plurality of second training data, and the second training data includes the third text content;
the processing unit 920 is further configured to: training a text model through the second training data set, wherein the text model comprises a semantic generation module and a text prediction module, a full connection layer in the text prediction module receives a first semantic vector and a plurality of second semantic vectors output by the semantic generation module, and the first semantic vector contains semantic information corresponding to the plurality of second semantic vectors; and after the loss function of the text model is converged to a preset value, generating the feature model, wherein the feature model comprises the semantic generation module and a feature prediction module, and a full connection layer in the feature prediction module receives a first semantic vector output by the semantic generation module.
Optionally, the processing unit 920 is specifically configured to: performing mask processing on at least one element in the third text content through a text model aiming at each second training data in the second training data set to obtain mask processed second training data, wherein the mask processed second training data comprises mask processed third text content and a content tag, and the content tag is at least one element replaced by a mask; training the text model through the second training data after the plurality of mask processing.
Fig. 10 is a schematic hardware structure of an electronic device 1000 according to an embodiment of the present application. As shown in fig. 10, generally, an electronic device 1000 includes: a processor 1001 and a memory 1002.
The processor 1001 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and so on. The processor 1001 may be implemented in at least one hardware form of DSP (Digital Signal Processing ), FPGA (Field-Programmable Gate Array, field programmable gate array), PLA (Programmable Logic Array ). The processor 1001 may also include a main processor, which is a processor for processing data in an awake state, also referred to as a CPU (Central Processing Unit ), and a coprocessor; a coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 1001 may integrate a GPU (Graphics Processing Unit, image processor) for rendering and drawing of content required to be displayed by the display screen. In some embodiments, the processor 1001 may also include an AI (Artificial Intelligence ) processor for processing computing operations related to machine learning.
Optionally, as shown in fig. 10, the electronic device 1000 may further include a transceiver 1003, and the processor 1001 may control the transceiver 1003 to communicate with other devices, and in particular, may send information or data to other devices, or receive information or data sent by other devices.
Wherein the transceiver 1003 may include a transmitter and a receiver. The transceiver 1003 may further include antennas, and the number of antennas may be one or more.
Optionally, the electronic device 1000 may implement corresponding flows in the methods of the embodiments of the present application, which are not described herein for brevity.
Those skilled in the art will appreciate that the structure shown in fig. 10 is not limiting of the electronic device 1000 and may include more or fewer components than shown, or may combine certain components, or may employ a different arrangement of components.
Embodiments of the present application also provide a non-transitory computer readable storage medium that, when executed by a processor of a node of a blockchain, enables the blockchain node to perform the method provided by the above embodiments.
The computer readable storage medium in this embodiment may be any available medium that can be accessed by a computer or a data storage device such as a server, a data center, etc. that contains one or more available medium(s) integrated, and the available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., an SSD), etc.
Those of ordinary skill in the art will appreciate that: all or part of the steps for implementing the method embodiments described above may be performed by hardware associated with program instructions. The foregoing program may be stored in a computer readable storage medium. The program, when executed, performs steps including the method embodiments described above; and the aforementioned storage medium includes: various media that can store program code, such as ROM, RAM, magnetic or optical disks.
The present embodiments also provide a computer program product comprising instructions which, when run on a computer, cause the computer to perform the method provided by the above embodiments.
It will be understood by those skilled in the art that all or part of the steps for implementing the above embodiments may be implemented by hardware, or may be implemented by a program for instructing relevant hardware, where the program may be stored in a computer readable storage medium, and the storage medium may be a read-only memory, a magnetic disk or an optical disk, etc.
The foregoing description of the preferred embodiments of the present application is not intended to limit the invention to the particular embodiments of the present application, but to limit the scope of the invention to the particular embodiments of the present application.
Claims (5)
1. A method for identifying a user feature, comprising:
acquiring a plurality of first text messages of a user, wherein the first text messages comprise first text contents and time stamps corresponding to the first text contents;
determining at least one feature of the user based on the plurality of first text messages;
outputting at least one characteristic of the user;
said determining at least one feature of said user from said plurality of first text messages comprises:
mapping, for each first text message, the first text message into a sequence of text vectors for characterizing the first text content and time information corresponding to the first text content in a vector space;
determining at least one feature of the user according to a plurality of text vector sequences corresponding to the plurality of first text information;
said determining at least one feature of said user from said plurality of first text messages comprises:
inputting the plurality of first text information into a pre-trained feature model to obtain at least one feature of the user;
before said inputting the plurality of first text information into the feature model to obtain at least one feature of the user, the method further comprises:
acquiring a first training data set, wherein the first training data set comprises a plurality of first training data, the first training data comprises a plurality of second text information and at least one characteristic label, and the second text information comprises second text content and a timestamp corresponding to the second text content;
training the feature model by the first training data set;
before the training of the feature model by the first training dataset, the method further comprises:
acquiring a second training data set, wherein the second training data set comprises a plurality of second training data, and the second training data comprises third text content;
training a text model through the second training data set, wherein the text model comprises a semantic generation module and a text prediction module, a full connection layer in the text prediction module receives a first semantic vector and a plurality of second semantic vectors output by the semantic generation module, and the first semantic vector contains semantic information corresponding to the plurality of second semantic vectors;
after the loss function of the text model is converged to a preset value, generating a feature model, wherein the feature model comprises a semantic generation module and a feature prediction module, and a full connection layer in the feature prediction module receives a first semantic vector output by the semantic generation module;
the training of the text model by the second training data set comprises:
performing mask processing on at least one element in the third text content through a text model aiming at each second training data in the second training data set to obtain mask processed second training data, wherein the mask processed second training data comprises mask processed third text content and a content tag, and the content tag is at least one element replaced by a mask;
training the text model through the second training data after the plurality of mask processing.
2. The method of claim 1, wherein said mapping said first text information into a sequence of text vectors for each first text information comprises:
mapping first text content of the first text information into a sequence of content vectors, the sequence of content vectors being used to characterize the first text content in a vector space;
mapping the timestamp corresponding to the first text content into a time vector sequence according to a preset time dimension, wherein the time dimension comprises at least one of year, month, day, week, hour or minute, and the time vector sequence is used for representing the time information of the first text content in the time dimension in a vector space;
and fusing the content vector sequence and the time vector sequence to obtain the text vector sequence.
3. An apparatus for identifying a user feature, comprising:
an acquisition unit configured to acquire a plurality of first text information of a user, the first text information including first text content and a timestamp corresponding to the first text content;
a processing unit for determining at least one feature of the user from the plurality of first text information;
an output unit for outputting at least one feature of the user;
the processing unit is further configured to:
mapping, for each first text message, the first text message into a sequence of text vectors for characterizing the first text content and time information corresponding to the first text content in a vector space;
determining at least one feature of the user according to a plurality of text vector sequences corresponding to the plurality of first text information;
the processing unit is further configured to:
inputting the plurality of first text information into a pre-trained feature model to obtain at least one feature of the user;
the processing unit further comprises a first training unit, the first training unit is used for acquiring a first training data set, the first training data set comprises a plurality of first training data, the first training data comprises a plurality of second text information and at least one characteristic label, and the second text information comprises second text content and a timestamp corresponding to the second text content;
training the feature model by the first training data set;
the first training unit is further preceded by a second training unit, the second training unit is used for acquiring a second training data set, the second training data set comprises a plurality of second training data, and the second training data comprises third text content;
training a text model through the second training data set, wherein the text model comprises a semantic generation module and a text prediction module, a full connection layer in the text prediction module receives a first semantic vector and a plurality of second semantic vectors output by the semantic generation module, and the first semantic vector contains semantic information corresponding to the plurality of second semantic vectors;
after the loss function of the text model is converged to a preset value, generating a feature model, wherein the feature model comprises a semantic generation module and a feature prediction module, and a full connection layer in the feature prediction module receives a first semantic vector output by the semantic generation module;
the second training unit is further configured to perform mask processing on at least one element in the third text content through a text model for each second training data in the second training data set to obtain mask processed second training data, where the mask processed second training data includes mask processed third text content and a content tag, and the content tag is at least one element replaced by a mask;
training the text model through the second training data after the plurality of mask processing.
4. An electronic device, comprising: a processor and a memory for storing a computer program, the processor being adapted to invoke and run the computer program stored in the memory for performing the method according to any of claims 1 to 2.
5. A computer readable storage medium storing a computer program for causing a computer to perform the method of any one of claims 1 to 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011281605.4A CN112364663B (en) | 2020-11-16 | 2020-11-16 | User characteristic identification method, device, equipment and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011281605.4A CN112364663B (en) | 2020-11-16 | 2020-11-16 | User characteristic identification method, device, equipment and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112364663A CN112364663A (en) | 2021-02-12 |
| CN112364663B true CN112364663B (en) | 2023-05-26 |
Family
ID=74515041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011281605.4A Active CN112364663B (en) | 2020-11-16 | 2020-11-16 | User characteristic identification method, device, equipment and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112364663B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110084067A (en) * | 2019-05-07 | 2019-08-02 | 欢动无限(北京)科技有限公司 | A kind of method for secret protection and device based on privacy chain |
| CN110490686A (en) * | 2019-07-08 | 2019-11-22 | 西北大学 | A kind of building of commodity Rating Model, recommended method and system based on Time Perception |
| CN111694949A (en) * | 2019-03-14 | 2020-09-22 | 京东数字科技控股有限公司 | Multi-text classification method and device |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109840321B (en) * | 2017-11-29 | 2022-02-01 | 腾讯科技（深圳）有限公司 | Text recommendation method and device and electronic equipment |
| CN110287961B (en) * | 2019-05-06 | 2024-04-09 | 平安科技（深圳）有限公司 | Chinese word segmentation method, electronic device and readable storage medium |
-
2020
- 2020-11-16 CN CN202011281605.4A patent/CN112364663B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111694949A (en) * | 2019-03-14 | 2020-09-22 | 京东数字科技控股有限公司 | Multi-text classification method and device |
| CN110084067A (en) * | 2019-05-07 | 2019-08-02 | 欢动无限(北京)科技有限公司 | A kind of method for secret protection and device based on privacy chain |
| CN110490686A (en) * | 2019-07-08 | 2019-11-22 | 西北大学 | A kind of building of commodity Rating Model, recommended method and system based on Time Perception |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112364663A (en) | 2021-02-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110288049B (en) | Method and apparatus for generating image recognition model | |
| US20180174037A1 (en) | Suggesting resources using context hashing | |
| CN111190939A (en) | User portrait construction method and device | |
| CN109086303A (en) | The Intelligent dialogue method, apparatus understood, terminal are read based on machine | |
| CN111967224A (en) | Method and device for processing dialog text, electronic equipment and storage medium | |
| CN110364146A (en) | Audio recognition method, device, speech recognition apparatus and storage medium | |
| CN110363084A (en) | A kind of class state detection method, device, storage medium and electronics | |
| CN110472154A (en) | A kind of resource supplying method, apparatus, electronic equipment and readable storage medium storing program for executing | |
| CN110096617B (en) | Video classification method and device, electronic equipment and computer-readable storage medium | |
| CN107943583A (en) | Processing method, device, storage medium and the electronic equipment of application program | |
| CN112801719A (en) | User behavior prediction method, user behavior prediction device, storage medium, and apparatus | |
| CN115114439B (en) | Method and device for multi-task model reasoning and multi-task information processing | |
| CN112650842A (en) | Human-computer interaction based customer service robot intention recognition method and related equipment | |
| CN110097004B (en) | Facial expression recognition method and device | |
| CN114418093B (en) | Method and device for training path characterization model and outputting information | |
| CN113919360A (en) | Semantic understanding method, voice interaction method, device, equipment and storage medium | |
| CN113641835B (en) | Multimedia resource recommendation method and device, electronic equipment and medium | |
| CN112364663B (en) | User characteristic identification method, device, equipment and storage medium | |
| CN112052316A (en) | Model evaluation method, model evaluation device, storage medium and electronic equipment | |
| CN112905787A (en) | Text information processing method, short message processing method, electronic device and readable medium | |
| CN116308757A (en) | Credit wind control model training method and device based on knowledge distillation, electronic equipment and computer medium | |
| CN111104951A (en) | Active learning method and device and terminal equipment | |
| CN111222011B (en) | Video vector determining method and device | |
| CN111488737B (en) | Text recognition method, device and equipment | |
| CN111967253A (en) | Entity disambiguation method and device, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 401121 b7-7-2, Yuxing Plaza, No.5 Huangyang Road, Yubei District, Chongqing Applicant after: Chongqing duxiaoman Youyang Technology Co.,Ltd. Address before: Room 3075, building 815, Jiayuan district, Shanghai Applicant before: SHANGHAI YOUYANG NEW MEDIA INFORMATION TECHNOLOGY Co.,Ltd. |
|
| CB02 | Change of applicant information | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20211217 Address after: Room 606, 6 / F, building 4, courtyard 10, Xibeiwang Road, Haidian District, Beijing 100085 Applicant after: Du Xiaoman Technology (Beijing) Co.,Ltd. Address before: 401121 b7-7-2, Yuxing Plaza, No.5 Huangyang Road, Yubei District, Chongqing Applicant before: Chongqing duxiaoman Youyang Technology Co.,Ltd. |
|
| TA01 | Transfer of patent application right | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |