CN117390577A - Feature acquisition method, device, apparatus, storage medium, and program product - Google Patents

Feature acquisition method, device, apparatus, storage medium, and program product Download PDF

Info

Publication number
CN117390577A
CN117390577A CN202311326921.2A CN202311326921A CN117390577A CN 117390577 A CN117390577 A CN 117390577A CN 202311326921 A CN202311326921 A CN 202311326921A CN 117390577 A CN117390577 A CN 117390577A
Authority
CN
China
Prior art keywords
behavior
tag
target
score
label
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.)
Pending
Application number
CN202311326921.2A
Other languages
Chinese (zh)
Inventor
徐程程
冯晓
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tencent Technology Shenzhen Co Ltd
Original Assignee
Tencent Technology Shenzhen Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tencent Technology Shenzhen Co Ltd filed Critical Tencent Technology Shenzhen Co Ltd
Priority to CN202311326921.2A priority Critical patent/CN117390577A/en
Publication of CN117390577A publication Critical patent/CN117390577A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90—Details of database functions independent of the retrieved data types
    • G06F16/95—Retrieval from the web
    • G06F16/953—Querying, e.g. by the use of web search engines
    • G06F16/9535—Search customisation based on user profiles and personalisation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/243—Classification techniques relating to the number of classes
    • G06F18/2431—Multiple classes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Databases & Information Systems (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The application provides a feature acquisition method, a device, equipment, a storage medium and a program product, which can be applied to various scenes such as cloud technology, artificial intelligence, intelligent traffic, auxiliary driving and the like; the method comprises the following steps: acquiring interactive content corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at popularization information; extracting tags from the interactive content to obtain a plurality of behavior tags of the target object; acquiring label scores of the behavior labels; selecting at least one target behavior label from the plurality of behavior labels based on label scores of the behavior labels, and taking the at least one target behavior label as a behavior characteristic of the target object, wherein the behavior characteristic is used for recommending popularization information for the target object; according to the method and the device, the recommending effect of recommending the popularization information based on the acquired object characteristics can be improved.

Description

Feature acquisition method, device, apparatus, storage medium, and program product
Technical Field
The present disclosure relates to the field of artificial intelligence, and in particular, to a feature acquisition method, apparatus, device, storage medium, and program product.
Background
Artificial intelligence (Artificial Intelligence, AI) is a comprehensive technology of computer science, and by researching the design principles and implementation methods of various intelligent machines, the machines have the functions of sensing, reasoning and decision. Artificial intelligence technology is a comprehensive subject, and relates to a wide range of fields, such as natural language processing technology, machine learning/deep learning and other directions, and with the development of technology, the artificial intelligence technology will be applied in more fields and has an increasingly important value.
The recommendation of promotion information is also an important application direction of artificial intelligence. In the related art, features of an object are generally extracted according to interactive behavior (e.g., clicking, viewing, etc.) performed by the object (e.g., a user) on promotional information, so that recommendation of promotional information to the object is performed based on the extracted features. However, the features of the object that can be extracted are limited, and the object cannot be known sufficiently, so that the recommendation effect of the popularization information is not good.
Disclosure of Invention
The embodiment of the application provides a feature acquisition method, a feature acquisition device, electronic equipment, a computer-readable storage medium and a computer program product, which can improve the recommendation effect of popularization information recommendation based on acquired object features.
The technical scheme of the embodiment of the application is realized as follows:
the embodiment of the application provides a feature acquisition method, which comprises the following steps:
acquiring interactive content corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at popularization information;
extracting tags from the interactive content to obtain a plurality of behavior tags of the target object;
acquiring label scores of the behavior labels;
wherein the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, the behavior feature being used to recommend promotional information to the target object;
and selecting at least one target behavior label from the plurality of behavior labels based on the label score of each behavior label, and taking the at least one target behavior label as the behavior characteristic of the target object.
The embodiment of the application also provides a feature acquisition device, which comprises:
the first acquisition module is used for acquiring interactive contents corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at popularization information;
the extraction module is used for extracting the tags of the interactive contents to obtain a plurality of behavior tags of the target object;
The second acquisition module is used for acquiring the label score of each behavior label;
wherein the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, the behavior feature being used to recommend promotional information to the target object;
the selecting module is used for selecting at least one target behavior label from the plurality of behavior labels based on the label score of each behavior label, and taking the at least one target behavior label as the behavior characteristic of the target object.
In the above scheme, the extraction module is further configured to perform classification processing of multiple classification dimensions on the interactive content, so as to obtain classification categories of the interactive content in each classification dimension; and taking the classification category of the interactive content in each classification dimension as a plurality of behavior labels of the target object.
In the above scheme, the extracting module is further configured to obtain a plurality of candidate tags; respectively carrying out score prediction on each candidate tag to obtain a prediction score of each candidate tag, wherein the prediction score indicates the possibility degree that the candidate tag is a behavior tag of the target object; and taking the plurality of candidate labels with the predictive scores meeting the predictive score condition as a plurality of behavior labels of the target object.
In the above scheme, the second obtaining module is further configured to obtain a tag score of each behavior tag in the current time period; the second obtaining module is further configured to obtain a first fractional attenuation parameter, and for each behavior tag, perform the following processing respectively: when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and taking the initial tag score as the tag score of the behavior tag in the current time period; and when the current time period is the j-th time period for determining the tag score, determining the tag score of the behavior tag in the current time period based on the first score attenuation parameter and the tag score of the behavior tag in the (j-1) -th time period, wherein j is an integer greater than 1.
In the above scheme, the second obtaining module is further configured to obtain a tag score of each behavior tag in the current time period; the second obtaining module is further configured to obtain a second fractional attenuation parameter, and for each behavior tag, perform the following processing respectively: when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and determining the tag score of the behavior tag in the current time period based on the initial tag score and the behavior indication parameter of the current time period; determining a tag score of the behavior tag in a current time period based on the second score decay parameter, the tag score of the behavior tag in the (i-1) th time period, the initial tag score, and the behavior indication parameter when the current time period is the i-th time period in which the tag score is determined; the behavior indication parameter indicates whether a target interaction behavior corresponding to the behavior tag exists in the current time period, and i is an integer greater than 1.
In the above scheme, the second obtaining module is further configured to determine a correlation degree between the behavior tag and the interactive content; and determining an initial tag score of the behavior tag based on the correlation degree corresponding to the behavior tag, wherein the initial tag score and the correlation degree are in a positive correlation relationship.
In the above scheme, the selecting module is further configured to determine, for each of the behavior tags, a behavior source corresponding to the behavior tag, and determine the occurrence number of the behavior tag in the behavior source; selecting first behavior tags with the number meeting a number condition from the behavior tags; determining a target tag score of a second behavior tag, which is other than the first behavior tag, according to the tag score of the second behavior tag and the behavior source weight of each behavior source corresponding to the second behavior tag, aiming at the second behavior tag in the plurality of behavior tags; selecting a third behavior label with the target label score meeting a score condition from the second behavior labels; and taking the first behavior label and the third behavior label as the at least one target behavior label.
In the above scheme, the selecting module is further configured to determine a behavior source corresponding to each behavior tag, and obtain a behavior source weight of each behavior source; determining a target tag score of each behavior tag based on the tag score of the behavior tag and the behavior source weight of each behavior source corresponding to the behavior tag; and selecting at least one fourth behavior label with the target label score meeting a score condition from the plurality of behavior labels, and taking the at least one fourth behavior label as the at least one target behavior label.
In the above solution, when the number of the target behavior tags is a plurality of, the selecting module is further configured to obtain, after the tag score based on each of the behavior tags selects at least one target behavior tag from the plurality of behavior tags, an interaction parameter value of a first object group for target popularization information, where the target popularization information carries the at least one target behavior tag, and the interaction parameter value indicates a possibility that the object group performs an interaction for the target popularization information; for each target behavior label, determining a second object group with the target behavior label, and determining a target interaction parameter value of the second object group for the target popularization information; selecting a fifth behavior label with the target interaction parameter value larger than the interaction parameter value from a plurality of target behavior labels; the selecting module is further configured to use the fifth behavior label as a behavior feature of the target object.
In the above scheme, when the number of the target behavior tags is multiple, the selecting module is further configured to, after selecting at least one target behavior tag from the multiple behavior tags based on the tag score of each of the behavior tags, perform score prediction on each of the target behavior tags to obtain a prediction score, where the prediction score indicates a correlation between the target behavior tag and a recommended target of popularization information; selecting a sixth behavior label with the predictive score meeting a score condition from a plurality of target behavior labels; the act of taking the at least one target act tag as the act feature of the target object includes: and taking the sixth behavior label as the behavior characteristic of the target object.
In the above scheme, the selecting module is further configured to obtain, after the at least one target behavior tag is used as a behavior feature of the target object, a matching relationship between the behavior feature and a promotion information feature, and obtain a promotion information feature to be recommended of each of a plurality of promotion information to be recommended; determining target popularization information characteristics matched with the behavior characteristics in the to-be-recommended popularization information characteristics based on the matching relation; and sending the popularization information to be recommended, which has the characteristics of the target popularization information, from the plurality of popularization information to be recommended to the terminal of the target object.
The embodiment of the application also provides electronic equipment, which comprises:
a memory for storing computer executable instructions;
and the processor is used for realizing the feature acquisition method provided by the embodiment of the application when executing the computer executable instructions stored in the memory.
The embodiment of the application also provides a computer readable storage medium, which stores computer executable instructions or a computer program, and when the computer executable instructions or the computer program are executed by a processor, the method for acquiring the characteristics provided by the embodiment of the application is realized.
The embodiment of the application also provides a computer program product, which comprises computer executable instructions or a computer program, and the computer executable instructions or the computer program realize the feature acquisition method provided by the embodiment of the application when being executed by a processor.
The embodiment of the application has the following beneficial effects:
according to the method and the device for recommending the popularization information, the interactive content corresponding to the target interactive behavior of the target object is firstly obtained, then the interactive content is subjected to label extraction to obtain a plurality of behavior labels of the target object, and then label scores of the behavior labels are obtained, and because the label scores indicate the possibility that the behavior labels are used as the behavior characteristics of the target object, at least one target behavior label can be selected from the plurality of behavior labels based on the label scores of the behavior labels, and therefore the at least one target behavior label is used as the behavior characteristics of the target object, and the popularization information is recommended to the target object based on the behavior characteristics.
Here, the behavior characteristics of the target object for promotion information recommendation are extracted from the interactive content corresponding to the target interactive behavior, and the target interactive behavior does not include the interactive behavior for promotion information. Therefore, the recommendation of the promotion information based on the interaction behavior of the non-promotion information domain is realized, the behavior characteristics of the object used for the recommendation of the promotion information are enriched, and when the promotion information is recommended to the object based on the acquired behavior characteristics of the object, the recommendation precision of the promotion information is higher, the selection range of the promotion information is wider, and the recommendation effect of the promotion information and the promotion effect of the promotion information are improved.
Drawings
FIG. 1 is a schematic architecture diagram of a feature acquisition system provided in an embodiment of the present application;
fig. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present application;
fig. 3 is a schematic flow chart of a feature acquisition method provided in an embodiment of the present application;
fig. 4 is a schematic diagram of a label extraction procedure provided in an embodiment of the present application;
fig. 5 is a schematic diagram of a label extraction procedure provided in an embodiment of the present application;
fig. 6 is a schematic diagram of a label score obtaining flow provided in an embodiment of the present application;
fig. 7 is a schematic diagram of a label score obtaining flow provided in an embodiment of the present application;
FIG. 8 is a schematic diagram of a selection process of a target behavior tag according to an embodiment of the present disclosure;
FIG. 9 is a schematic diagram of a selection flow of a target behavior tag according to an embodiment of the present application;
fig. 10 is a schematic diagram of a screening process of target behavior tags according to an embodiment of the present application.
Detailed Description
For the purpose of making the objects, technical solutions and advantages of the present application more apparent, the present application will be described in further detail with reference to the accompanying drawings, and the described embodiments should not be construed as limiting the present application, and all other embodiments obtained by those skilled in the art without making any inventive effort are within the scope of the present application.
In the following description, reference is made to "some embodiments" which describe a subset of all possible embodiments, but it is to be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments and can be combined with one another without conflict.
In the following description, the terms "first", "second", "third" and the like are merely used to distinguish similar objects and do not represent a specific ordering of the objects, it being understood that the "first", "second", "third" may be interchanged with a specific order or sequence, as permitted, to enable embodiments of the application described herein to be practiced otherwise than as illustrated or described herein.
Unless defined otherwise, all technical and scientific terms used in the embodiments of the present application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in the embodiments of the application is for the purpose of describing the embodiments of the application only and is not intended to be limiting of the application.
Before further describing embodiments of the present application in detail, the terms and expressions that are referred to in the embodiments of the present application are described, and are suitable for the following explanation.
1) Client side: applications running in the terminal for providing various services, such as a client supporting promotion information recommendation.
2) In response to: for representing a condition or state upon which an operation is performed, one or more operations performed may be in real-time or with a set delay when the condition or state upon which the operation is dependent is satisfied; without being specifically described, there is no limitation in the execution sequence of the plurality of operations performed.
3) The two-party behavior (i.e., target interaction behavior) is a behavior which is not directly related to the behavior of the user aiming at the popularization information (such as clicking the popularization information, browsing the popularization information, purchasing an object promoted by the popularization information through the popularization information, adding a shopping cart, downloading and the like), namely, the two-party behavior is not a behavior aiming at the popularization information field, such as a searching behavior, a video watching behavior, a reading behavior aiming at an article or a novel and the like.
Based on the above description of the terms and terminology involved in the embodiments of the present application, the embodiments of the present application will be described in detail below. The embodiment of the application provides a feature acquisition method, a feature acquisition device, electronic equipment, a computer-readable storage medium and a computer program product, which can improve the recommendation effect of popularization information recommendation based on acquired object features.
It should be noted that, in the application of the present application, the relevant data collection process should strictly obtain the informed consent or the individual consent of the personal information body according to the requirements of the relevant laws and regulations, and develop the subsequent data use and processing actions within the authorized range of the laws and regulations and the personal information body.
The following describes a feature acquisition system provided in an embodiment of the present application. Referring to fig. 1, fig. 1 is a schematic architecture diagram of a feature acquisition system provided in an embodiment of the present application. To enable support for one exemplary application, the feature acquisition system 100 includes: server 200, network 300, and terminal 400. The terminal 400 is connected to the server 200 through the network 300, where the network 300 may be a wide area network or a local area network, or a combination of both, and the data transmission is implemented using a wireless or wired link.
Here, the terminal 400 (e.g., a client running with support for promotion information recommendation) transmits a recommendation request of promotion information to the server 200 in response to a view instruction of a target object for a page including promotion information; the server 200 receives a recommendation request of promotion information sent by the terminal 400; responding to a recommendation request of popularization information, acquiring interactive content corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at the popularization information; extracting tags of the interactive contents to obtain a plurality of behavior tags of the target object; acquiring tag scores of the behavior tags, wherein the tag scores indicate the possibility of the behavior tags serving as behavior characteristics of the target object; selecting at least one target behavior label from a plurality of behavior labels based on label scores of the behavior labels, and taking the at least one target behavior label as a behavior characteristic of a target object; determining target popularization information recommended to the target object from a plurality of pieces of popularization information to be recommended based on behavior characteristics of the target object; sending target popularization information to the terminal 400; the terminal 400 receives the target popularization information returned by the server 200; and displaying the target popularization information in the page.
In some embodiments, the feature acquisition method provided in the embodiments of the present application is implemented by an electronic device, for example, may be implemented by a terminal alone, may be implemented by a server alone, or may be implemented by a terminal and a server cooperatively. The embodiments of the present application may be applied to various scenarios including, but not limited to, cloud technology, artificial intelligence, intelligent transportation, assisted driving, audio-visual, instant messaging, gaming, etc.
In some embodiments, the electronic device implementing the feature acquisition method provided in the embodiments of the present application may be various types of terminals or servers. The server (e.g., server 200) may be an independent physical server, or may be a server cluster or a distributed system formed by a plurality of physical servers. The terminal (e.g., terminal 400) may be, but is not limited to, a notebook computer, tablet computer, desktop computer, smart phone, smart voice interaction device (e.g., smart speaker), smart home appliance (e.g., smart television), smart watch, vehicle-mounted terminal, wearable device, virtual Reality (VR) device, aircraft, etc. The terminal and the server may be directly or indirectly connected through wired or wireless communication, which is not limited in the embodiment of the present application.
In some embodiments, the feature acquisition method provided by the embodiments of the present application may be implemented by means of Cloud Technology (Cloud Technology). Cloud technology refers to a hosting technology for unifying serial resources such as hardware, software, network and the like in a wide area network or a local area network to realize calculation, storage, processing and sharing of data. The cloud technology is a generic term of network technology, information technology, integration technology, management platform technology, application technology and the like based on cloud computing business model application, can form a resource pool, and is flexible and convenient as required. Cloud computing technology will become an important support. Background services of technical network systems require a large amount of computing and storage resources. As an example, a server (e.g., server 200) may also be a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, web services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDNs), and basic cloud computing services such as big data and artificial intelligence platforms.
In some embodiments, the feature acquisition method provided in the embodiments of the present application may be implemented by means of a Block Chain (Block Chain) technology. Blockchains are novel application modes of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanisms, encryption algorithms, and the like. By way of example, multiple servers may be organized into a blockchain, and the servers may be nodes on the blockchain, with information connections between each node in the blockchain, and information transfer between the nodes may be via the information connections. The data (e.g., the behavior characteristics of the target object) related to the feature acquisition method provided in the embodiments of the present application may be stored on a blockchain.
In some embodiments, the terminal or server may implement the feature acquisition method provided in the embodiments of the present application by running various computer-executable instructions or computer programs. For example, the computer-executable instructions may be commands at the micro-program level, machine instructions, or software instructions. The computer program may be a native program or a software module in an operating system; a local (Native) Application (APP), i.e. a program that needs to be installed in an operating system to run, for example, an APP that supports promotion information recommendation; or an applet that can be embedded in any APP, i.e., a program that can be run only by being downloaded into the browser environment. In general, the computer-executable instructions may be any form of instructions and the computer program may be any form of application, module, or plug-in.
The electronic device for implementing the feature acquisition method provided in the embodiment of the present application is described below. Referring to fig. 2, fig. 2 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device 500 provided in the embodiment of the present application may be a terminal or a server. As shown in fig. 2, the electronic device 500 includes: at least one processor 510, a memory 550, at least one network interface 520, and a user interface 530. The various components in electronic device 500 are coupled together by bus system 540. It is appreciated that the bus system 540 is used to enable connected communications between these components. The bus system 540 includes a power bus, a control bus, and a status signal bus in addition to the data bus. The various buses are labeled as bus system 540 in fig. 2 for clarity of illustration.
The processor 510 may be an integrated circuit chip with signal processing capabilities such as a general purpose processor, such as a microprocessor or any conventional processor, a digital signal processor (Digital Signal Processor, DSP), or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or the like.
The user interface 530 includes one or more output devices 531 that enable presentation of media content, including one or more speakers and/or one or more visual displays. The user interface 530 also includes one or more input devices 532, including user interface components that facilitate user input, such as a keyboard, mouse, microphone, touch screen display, camera, other input buttons and controls.
The memory 550 may be removable, non-removable, or a combination thereof. Memory 550 may include one or more storage devices physically located away from processor 510. Memory 550 includes volatile memory or nonvolatile memory, and may also include both volatile and nonvolatile memory. The nonvolatile Memory may be a Read Only Memory (ROM), and the volatile Memory may be a random access Memory (Random Access Memory, RAM). The memory 550 described in embodiments herein is intended to comprise any suitable type of memory.
In some embodiments, memory 550 is capable of storing data to support various operations, examples of which include programs, modules and data structures, or subsets or supersets thereof, as exemplified below.
An operating system 551 including system programs for handling various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and handling hardware-based tasks;
network communication module 552 is used to reach other electronic devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 include: bluetooth, wireless compatibility authentication (WiFi), and universal serial bus (Universal Serial Bus, USB), etc.;
a presentation module 553 for enabling presentation of information (e.g., a user interface for operating a peripheral device and displaying content and information) via one or more output devices 531 (e.g., a display screen, speakers, etc.) associated with the user interface 530;
the input processing module 554 is configured to detect one or more user inputs or interactions from one of the one or more input devices 532 and translate the detected inputs or interactions.
In some embodiments, the feature acquiring apparatus provided in the embodiments of the present application may be implemented in a software manner, and fig. 2 shows the feature acquiring apparatus 555 stored in the memory 550, which may be software in the form of a program, a plug-in, or the like, including the following software modules: the first acquisition module 5551, the extraction module 5552, the second acquisition module 5553 and the selection module 5554 are logical, and thus may be arbitrarily combined or further split according to the functions implemented, the functions of each module will be described below.
The feature acquisition method provided in the embodiment of the present application is described below. As described above, the feature acquiring method provided in the embodiments of the present application is implemented by an electronic device, for example, may be implemented by a server or a terminal alone, or implemented by the server and the terminal cooperatively. The execution subject of each step will not be repeated hereinafter. Referring to fig. 3, fig. 3 is a flowchart of a feature acquisition method provided in an embodiment of the present application, where the feature acquisition method provided in the embodiment of the present application includes:
step 101: and acquiring interactive content corresponding to the target interactive behavior of the target object.
Wherein the target interaction behavior does not include interaction behavior for promotional information.
In step 101, the target object may be a user; the target interaction behavior is a two-party behavior, and the two-party behavior is a behavior which is not in direct relation with the behavior of a user aiming at popularization information (such as clicking popularization information, browsing popularization information, purchasing, shopping cart adding, downloading and the like of an object promoted by the popularization information through the popularization information), namely the two-party behavior is not a behavior aiming at the field of the popularization information, for example, the two-party behavior can be a search behavior, a video watching behavior, a reading behavior aiming at an article or a novel and the like. The target interaction behavior often corresponds to the corresponding interaction content, i.e. the object is performing the target interaction behavior for the interaction content. For example, the target interaction behavior may be a search behavior, a video viewing behavior, a reading behavior for articles or novels, a use behavior for APPs, etc., and then the corresponding interaction content may be search content, video, articles, novels, APPs, etc., respectively. The number of the target interactions may be one or more, and the number of the interactions may be one or more.
Step 102: and extracting the tags of the interactive contents to obtain a plurality of behavior tags of the target object.
In step 102, for the interactive content corresponding to the obtained target interactive behavior, extracting a tag from the interactive content, where the tag extraction is to understand the content of the interactive content, so as to obtain a content tag of each interactive content. Because the object performs the target interaction behavior on the interaction content with the content tag, it can also be understood that the target interaction behavior of the object on the interaction content is the target interaction behavior of the object on the content tag, and therefore, the content tag can be understood as the behavior tag of the object. Therefore, after extracting the tags of the interactive contents and obtaining a plurality of content tags, the obtained plurality of content tags are taken as a plurality of behavior tags of the target object. Thus, a plurality of behavior tags of the target object are obtained by extracting tags of the interactive content.
In some embodiments, referring to fig. 4, fig. 4 shows that step 102 of fig. 3 may be implemented by: step 1021a, performing classification processing of multiple classification dimensions on the interactive content to obtain classification categories of the interactive content in each classification dimension; step 1022a, using the classification category of the interactive content in each classification dimension as a plurality of behavior tags of the target object.
Here, in step 1021a, the interactive content may be subjected to classification processing in a plurality of classification dimensions, so as to obtain classification categories of the interactive content in each classification dimension. The classification process may be performed by a classification model, which may be pre-built (e.g., built based on neural networks) and trained; one classification model may support classification processing of only one classification dimension, or may support classification processing of a plurality of classification dimensions simultaneously. In step 1022a, the classification category of the interactive content in each classification dimension may be used as a plurality of behavior tags of the target object.
In some embodiments, referring to fig. 5, fig. 5 shows that step 102 of fig. 3 may also be implemented by: step 1021b, obtaining a plurality of candidate tags; step 1022b, respectively performing score prediction on each candidate tag to obtain a prediction score of each candidate tag, where the prediction score indicates a possible degree of the candidate tag being a behavior tag of the target object; step 1023b, using the plurality of candidate labels with the prediction scores meeting the prediction score condition as a plurality of behavior labels of the target object.
Here, in step 1021b, a plurality of candidate tags may be acquired, which may be preset, such as comedy-type videos, electronic device recommendation-type videos, social-type APPs, tool-type APPs, and the like. In step 1022b, score prediction may be performed on each candidate tag to obtain a prediction score of each candidate tag, where the prediction score is used to indicate a likelihood of the candidate tag being a behavioral tag of the target object, and the prediction score and the likelihood are in a positive correlation relationship; the score prediction may be achieved by pre-building (e.g., based on neural networks) and training a completed score prediction model. In step 1023b, based on the obtained prediction scores of the candidate tags, a plurality of candidate tags whose prediction scores satisfy the prediction score condition may be selected from the plurality of candidate tags, so that the selected plurality of candidate tags are taken as a plurality of behavior tags of the target object. The predictive score condition may be preset. For example, a target number of candidate tags whose predictive scores are ranked in descending order may be selected as the behavior tags, and candidate tags whose predictive scores reach a predictive score threshold (which may be preset) may be selected as the behavior tags.
Step 103: and obtaining the label score of each behavior label.
The tag score indicates the possibility that the behavior tag is taken as the behavior characteristic of the target object, and the behavior characteristic is used for recommending the popularization information to the target object.
In step 103, after extracting a plurality of behavior tags of the target object, a tag score of each behavior tag is obtained, where the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, and the greater the tag score, the higher the likelihood. In the embodiment of the present application, the extracted behavior tags may represent the behavior features of the target object, so it is necessary to obtain the tag score of each behavior tag, so as to indicate the likelihood that the behavior tag is the behavior feature of the target object through the tag score, so as to determine which behavior tags are taken as the behavior feature of the target object based on the tag score. Here, the behavior feature is used for recommending the popularization information to the target object based on the behavior feature, so that the recommendation of the popularization information based on the interaction behavior of the non-popularization information domain is realized.
In some embodiments, referring to fig. 6, fig. 6 shows that step 103 of fig. 3 may be implemented by: step 1031, obtaining a first fractional attenuation parameter, and for each behavior label, executing the following processing respectively: step 1031a, determining whether the current time period is the 1 st time period of determining the tag score, if yes, executing step 1031b, otherwise, executing step 1031c; step 1031b, obtaining an initial tag score of the behavior tag, and taking the initial tag score as the tag score of the behavior tag in the current time period; in step 1031c, when the current time period is the jth time period for which the tag score is determined, the tag score of the behavior tag in the current time period is determined based on the first score decay parameter and the tag score of the behavior tag in the (j-1) th time period, j being an integer greater than 1.
Here, the tag score of each behavior tag is calculated periodically, and each period is a period of time (e.g., 1 day, 12 hours, etc.), that is, every other period of time, the tag score of each behavior tag is recalculated once. Therefore, when the tag score of each behavior tag is acquired, the tag score of each behavior tag in the current time period is acquired. Since the interaction content corresponding to the target interaction behavior of a period of time is the interaction content, the behavior label of the target object will change, and the recent behavior label is more important than the previous behavior label for determining the behavior characteristic of the current time point, therefore, in the embodiment of the application, the label score of the previous behavior label should be gradually attenuated, so that the first score attenuation parameter of the label score of the behavior label is set, and the value of the first score attenuation parameter is between 0 and 1. Therefore, the determined tag score can be more accurate, and the behavior tag representing the recent behavior feature of the target object can be accurately selected according to the tag score.
In step 1031, the following processing may be performed for each behavior label to obtain a label score of each behavior label: in step 1031a, it is first determined whether the current time period is the 1 st time period for which the tag score is determined. If yes, in step 1031b, an initial tag score of the behavior tag is obtained, and the initial tag score is used as the tag score of the behavior tag in the current time period. If not, in step 1031c, when the current time period is the jth time period for determining the tag score, determining the tag score of the behavior tag in the current time period based on the first score attenuation parameter and the tag score of the behavior tag in the (j-1) th time period, that is: label score of behavioral label at jth time period = first score decay parameter the label score of behavioral label at (j-1) th time period.
In some embodiments, referring to fig. 7, fig. 7 shows that step 103 of fig. 3 may also be implemented by: step 1032, obtaining the second fractional attenuation parameters, and for each behavior label, performing the following processing respectively: step 1032a, determining whether the current time period is the 1 st time period of determining the tag score, if yes, executing step 1032b, otherwise, executing step 1032c; step 1032b, obtaining an initial tag score of the behavior tag, and determining a tag score of the behavior tag in the current time period based on the initial tag score and the behavior indication parameter of the current time period; step 1032c, when the current time period is the i-th time period for which the tag score is determined, determining the tag score of the behavior tag in the current time period based on the second score attenuation parameter, the tag score of the behavior tag in the (i-1) -th time period, the initial tag score, and the behavior indication parameter; the behavior indication parameter indicates whether a target interaction behavior corresponding to a behavior label exists in the current time period, and i is an integer greater than 1.
Here, the tag score of each behavior tag is calculated periodically, and each period is a period of time (e.g., 1 day, 12 hours, etc.), that is, every other period of time, the tag score of each behavior tag is recalculated once. Therefore, when the tag score of each behavior tag is acquired, the tag score of each behavior tag in the current time period is acquired. Since the interaction content corresponding to the target interaction behavior of the target object is a period of time, the behavior label of the target object will change, and the recent behavior label is more important than the previous behavior label for determining the behavior feature of the current time point, therefore, in the embodiment of the present application, the label score of the previous behavior label should be gradually attenuated, so that the second score attenuation parameter of the label score of the behavior label is set here, and the value of the second score attenuation parameter is between 0 and 1. Therefore, the determined tag score can be more accurate, and the behavior tag representing the recent behavior feature of the target object can be accurately selected according to the tag score.
In step 1032, the following processing may be performed for each behavior label to obtain a label score of each behavior label: in step 1032a, it is first determined whether the current time period is the 1 st time period for which the tag score is determined. If yes, in step 1032b, an initial tag score of the behavior tag is obtained, and the tag score of the behavior tag in the current time period is determined based on the initial tag score and the behavior indication parameter of the current time period. Specifically, the tag score of the behavior tag at the current time period = behavior indication parameter. If not, in step 1032c, when the current time period is the i-th time period for which the tag score is determined, the tag score of the behavior tag in the current time period is determined based on the second score decay parameter, the tag score of the behavior tag in the (i-1) -th time period, the initial tag score, and the behavior indication parameter. Specifically, the label score of the behavior label in the i-th period=the second score decay parameter the label score of the behavior label in the (i-1) -th period+the behavior indication parameter the initial label score. It should be noted that, the behavior indication parameter indicates whether the target interaction behavior corresponding to the behavior label exists in the current time period.
In some embodiments, the initial tag score of the behavior tag may be obtained by: determining the relativity of the behavior label and the interactive content; and determining an initial tag score of the behavior tag based on the correlation degree corresponding to the behavior tag, wherein the initial tag score and the correlation degree are in a positive correlation relationship. Here, since the behavior tag is actually a content tag of the interactive content, an initial tag score of the behavior tag may be determined based on a degree of correlation of the content tag with the interactive content. The correlation calculation may be implemented by Pearson correlation coefficient (Pearson), spearman correlation coefficient (Spearman's rank correlation coefficient), or the like. Different correlations correspond to different initial tag scores, and the initial tag scores and the correlations are positive correlations.
Step 104: and selecting at least one target behavior label from the plurality of behavior labels based on the label scores of the behavior labels, and taking the at least one target behavior label as the behavior characteristic of the target object.
In step 104, at least one target behavior tag from the plurality of behavior tags may be selected as a behavior feature of the target object based on the determined tag score of each behavior tag. For example, a behavior tag whose tag score reaches a tag score threshold may be selected as the target behavior tag; the behavior tags of the target number with the tag scores sorted in descending order can be selected as the target behavior tags. However, in practical applications, the target interaction behavior corresponding to the interaction content of the behavior tag is extracted and obtained often from a plurality of behavior sources, where the behavior sources characterize the source of executing the target interaction behavior, i.e. from which source the object enters and executes the target interaction behavior, and the behavior sources may be, for example, search entries, public numbers, content recommendation pages, and so on. Thus, the selection of behavior tags for different behavior sources can be achieved as follows:
In some embodiments, when there are a plurality of behavior sources corresponding to the behavior tags, referring to fig. 8, based on the tag score of each behavior tag, at least one target behavior tag may be selected from the plurality of behavior tags by: step 301, determining behavior sources corresponding to the behavior tags, and obtaining the behavior source weights of the behavior sources; step 302, determining a target tag score of each behavior tag based on the tag score of the behavior tag and the behavior source weight of each behavior source corresponding to the behavior tag; step 303, selecting at least one fourth behavior label with the target label score meeting the score condition from the plurality of behavior labels, and taking the at least one fourth behavior label as the at least one target behavior label.
Here, in step 301, the behavior sources corresponding to the behavior tags are first determined, and then the behavior source weights of the behavior sources are acquired. The behavior source corresponding to the behavior label is the behavior source of the target interaction behavior corresponding to the interaction content with the behavior label. The behavior source weights may be preset, with different behavior sources having different behavior source weights. In step 302, the following processing may be performed for each behavior tag to obtain a target tag score of each behavior tag: and weighting the tag scores of the behavior tags based on the behavior source weights of the behavior sources of the behavior tags to obtain target tag scores of the behavior tags. Thus, in step 303, at least one fourth behavior tag whose target tag score satisfies the score condition may be selected from the plurality of behavior tags based on the target tag score of each behavior tag, and the at least one fourth behavior tag is used as the at least one target behavior tag. Specifically, a fourth behavior label of a target number, which is ranked in descending order by the target label score, can be selected as a target behavior label; and a fourth behavior label with the target label score reaching a label score threshold can be selected as the target behavior label. In this way, the accuracy of selecting the target behavior tags can be improved by weighting the multi-source behavior tags.
In some embodiments, when there are a plurality of behavior sources corresponding to the behavior tags, referring to fig. 9, based on the tag score of each behavior tag, at least one target behavior tag may be selected from the plurality of behavior tags by: step 401, determining, for each behavior label, a behavior source corresponding to the behavior label, and determining the occurrence number of the behavior label in the behavior source; step 402, selecting the first behavior tags with the number meeting the number condition from a plurality of behavior tags; step 403, determining, for a second behavior tag other than the first behavior tag in the plurality of behavior tags, a target tag score of the second behavior tag based on a tag score of the second behavior tag and a behavior source weight of each behavior source corresponding to the second behavior tag; step 404, selecting a third behavior label with the target label score meeting the score condition from the second behavior labels; step 405, the first behavior tag and the third behavior tag are used as at least one target behavior tag.
Since weighting only the multisource behavior tags directly may ignore behavior tags for behaviors that are not highly weighted, the behavior tags may be current real behavior features of the object. Therefore, here, first, in step 401, for each behavior tag, at least one behavior source corresponding to the behavior tag is determined; since the at least one behavior source may appear the behavior tag multiple times (e.g., the interactive content of multiple target interactive behaviors of the behavior source may all have the behavior tag), the number of occurrences of the behavior tag at the behavior source is also determined; then, in step 402, selecting the first behavior tags whose number of occurrences satisfies the number condition from the plurality of behavior tags; for example, selecting a first behavior tag that presents a target number (e.g., 80% of the total number of behavior tags) that is ranked first in descending order of number; or selecting the first behavior label with the occurrence number reaching the number threshold value. After the first behavior tags whose number satisfies the number condition are selected from the plurality of behavior tags, in step 403, the tag score of the second behavior tag may be weighted based on the behavior source weights of the behavior sources corresponding to the second behavior tag with respect to the remaining unselected second behavior tags, so as to obtain the target tag score of the second behavior tag. Thus, in step 404, from the second behavior tags, a third behavior tag whose target tag score satisfies the score condition is selected, for example, a third behavior tag whose target tag score is ranked in descending order of the first target number may be selected; a third behavior tag may also be selected for which the target tag score reaches a tag score threshold. Finally, in step 405, the selected first behavior tag and third behavior tag are used as at least one target behavior tag. Thus, the target behavior label representing the behavior characteristic of the target object can be accurately selected.
In some embodiments, referring to fig. 10, when the number of target behavior tags is a plurality, after selecting at least one target behavior tag from the plurality of behavior tags, the at least one target behavior tag may be further filtered by: step 501, obtaining an interaction parameter value of a first object group aiming at target popularization information, wherein the target popularization information carries at least one target behavior label, and the interaction parameter value indicates the possibility of the object group to execute interaction behaviors aiming at the target popularization information; step 502, determining a second object group with target behavior labels according to each target behavior label, and determining target interaction parameter values of the second object group according to target popularization information; step 503, selecting a fifth behavior label with a target interaction parameter value greater than the interaction parameter value from the plurality of target behavior labels. Accordingly, when at least one target behavior tag is used as the behavior feature of the target object, the fifth behavior tag may be used as the behavior feature of the target object.
Here, since not all the target behavior tags are related to the recommendation of the promotional information, for example, if it is the latest spring festival, it is possible that most users have behavior tags such as "spring festival", "past year", and so the users are indistinguishable, and thus, the target behavior tags need to be filtered. In step 501, interaction parameter values of a first object group for target popularization information are obtained. The target popularization information carries the label of the same system as the behavior label, namely the target popularization information is equivalent to carrying at least one target behavior label; the first population of objects may be randomly selected; the interaction parameter value indicates the possibility of the object group to execute the interaction action aiming at the target promotion information, for example, the interaction parameter value can be click rate, conversion rate, click rate, conversion rate and the like, and the higher the interaction parameter value is, the better the promotion effect of the promotion information is. In step 502, a portion of users having each target behavior label, i.e., a second population of objects, may be determined from each target behavior label; and then determining a target interaction parameter value of the second object group aiming at the target popularization information, wherein the target interaction parameter value corresponds to the interaction parameter value, namely, the interaction parameter value is which of click rate, conversion rate and click rate is the conversion rate, and the target interaction parameter value is the same. In step 503, selecting a fifth behavior label with a target interaction parameter value greater than the interaction parameter value from the plurality of target behavior labels; here, the target interaction parameter value is greater than the interaction parameter value, which indicates that the corresponding target behavior label has significance for distinguishing users. So that the fifth behavioral label can be taken as the behavioral characteristic of the target object.
In some embodiments, when the number of target behavior tags is a plurality, after selecting at least one target behavior tag from the plurality of behavior tags, the at least one target behavior tag may be further filtered by: aiming at each target behavior label, carrying out score prediction on the target behavior label to obtain a prediction score, wherein the prediction score indicates the correlation between the target behavior label and a recommended target of popularization information; and selecting a sixth behavior label with the predictive score meeting the score condition from the plurality of target behavior labels. Accordingly, when at least one target behavior tag is used as the behavior feature of the target object, the sixth behavior tag may be used as the behavior feature of the target object.
Here, a score prediction model for score prediction may be previously constructed and trained, so that score prediction is performed on each target behavior label based on the score prediction model, and a prediction score of each target behavior label is obtained, so that the correlation between the target behavior label and the recommended target of the popularization information is indicated by the prediction score, and the prediction score and the correlation are positive correlations. Since not all target behavior tags are related to the recommendation of the promotional information, the target behavior tags need to be screened, namely: and selecting a sixth behavior label with the predictive score meeting the score condition from the plurality of target behavior labels. For example, a sixth behavior tab may be selected for which the predictive score reaches a predictive score threshold; a sixth behavior tab may also be selected that predicts a top target number in descending order of scores, which may be set as desired. So that the sixth behavior tag can be taken as the behavior feature of the target object.
In some embodiments, the recommendation of promotional information may be based on behavioral characteristics of the target object by: acquiring a matching relation between behavior characteristics and promotion information characteristics, and acquiring promotion information characteristics to be recommended of each promotion information to be recommended in a plurality of promotion information to be recommended; determining target popularization information characteristics matched with the behavior characteristics in the to-be-recommended popularization information characteristics based on the matching relation; and sending the popularization information to be recommended, which has the characteristics of the target popularization information, from among the plurality of popularization information to be recommended to the terminal of the target object.
According to the method and the device for recommending the popularization information, the interactive content corresponding to the target interactive behavior of the target object is firstly obtained, then the interactive content is subjected to label extraction to obtain a plurality of behavior labels of the target object, and then label scores of the behavior labels are obtained, and because the label scores indicate the possibility that the behavior labels are used as the behavior characteristics of the target object, at least one target behavior label can be selected from the plurality of behavior labels based on the label scores of the behavior labels, and therefore the at least one target behavior label is used as the behavior characteristics of the target object, and the popularization information is recommended to the target object based on the behavior characteristics.
Here, the behavior characteristics of the target object for promotion information recommendation are extracted from the interactive content corresponding to the target interactive behavior, and the target interactive behavior does not include the interactive behavior for promotion information. Therefore, the recommendation of the promotion information based on the interaction behavior of the non-promotion information domain is realized, the behavior characteristics of the object used for the recommendation of the promotion information are enriched, and when the promotion information is recommended to the object based on the acquired behavior characteristics of the object, the recommendation precision of the promotion information is higher, the selection range of the promotion information is wider, and the recommendation effect of the promotion information and the promotion effect of the promotion information are improved.
An exemplary application of the embodiments of the present application in a practical application scenario is described below.
In the recommended scenario of the popularization information, the two-party behaviors of the user (namely, the target interaction behaviors, the two-party behaviors and the behaviors (such as clicking and browsing) of the user aiming at the popularization information are not in direct relation, namely, the two-party behaviors are not behaviors aiming at the popularization information field, such as searching behaviors, video watching behaviors, reading behaviors aiming at articles or novels and the like), and the relation between the two-party behaviors and clicking and converting targets of the popularization information is indirect, but the two-party behaviors can reflect the real characteristics or intentions of the user. Therefore, the embodiment of the application provides a feature acquisition method, which can perform feature processing on the two-party behaviors of a user to obtain the two-party behavior features (namely, the behavior features), so that the popularization information is recommended through the obtained two-party behavior features, and the popularization effect of the popularization information is improved. Specifically, the feature acquisition method provided by the embodiment of the application comprises a time attenuation processing flow, a multi-source fusion processing flow and a tag filtering processing flow, so that the processed two-way behavior feature can reflect the real feature of a user, and the method has a gain on a recommendation system of popularization information; when the processed two-party behavior feature is applied to a recommendation system of popularization information, the weight of the two-party behavior feature is improved through two modes of feature cross processing and two-party behavior feature sub-network modeling, so that the conversion of the popularization information is better promoted.
First, a feature processing section in the embodiment of the present application is described. Here, the two-party behavior feature comes from the two-party behavior of the user (which may be a two-party behavior of a target time window (e.g., one month)), such as the behavior of the user reading an article, watching a video, the user searching a query, the user using a client, and the like. Therefore, in the feature processing of the two-party behavior, content (for example, an article, a video, a query, a client, etc.) related to the two-party behavior is first subjected to content understanding, and the content understanding may be to extract tags (tags) from the content. The label extraction can be 1) classifying the content, and taking the classification result as the content label of the corresponding content; 2) A tag set is acquired, each tag of the tag set is then scored for each content, and a certain number of tags with the scores obtained by the scoring being ranked in descending order are used as content tags of the content, wherein the score of each tag indicates the possibility that the tag is the content tag of the content. Then for each content tag of the content, determining an initial tag score of the content tag according to the relevance of the content tag and the content, wherein the initial tag score and the relevance are positively correlated. In this way, a content tag is obtained for each content involved in the two-party behavior, and each content tag has a corresponding initial tag score. In the embodiment of the present application, the behavior of the user on the content is also mapped to the behavior of the user on the content tag of the content, so the processing of the behavior features of the two parties includes the following processing flows:
(1) Time decay process flow. In some examples, when calculating the tag scores of the content tags for the current time period (the time period may be 1 day, 12 hours, etc.), the tag scores of the content tags involved in the user's two-party behavior during the previous time period are accumulated, but there are two problems: 1) The accumulation of tag scores is linear, the value is larger and larger, and the tag score of the content tag with longer time is high; 2) The user's content tags are time efficient, recent content tags are more important than older content tags, and linear overlays cannot highlight recent content tags. Accordingly, the above problems can be solved as follows:
for 1): in calculating the tag score of the content tag of the i-th time period, the tag score of the content tag of the (i-1) -th time period is multiplied by an attenuation factor α (having a value between 0 and 1, i.e., the above-mentioned second-score attenuation parameter). Thus, the tag score of the previous content tag can be reduced every time period, specifically expressed by the formula (one):
score i ＝α×score i-1 +C×weight，(0<α<1) The method comprises the steps of carrying out a first treatment on the surface of the Formula 1
Wherein, C (i.e. the behavior indication parameter) represents whether the current time period has the two-party behavior corresponding to the content tag, and the value is 0 (representing no two-party behavior) and 1 (representing two-party behavior); weight is the initial tag score; i represents an i-th period; score i Score representing tag score of content tag at i-th time period i-1 A tag score representing the content tag at the (i-1) th time period.
For 2): the label score of the current time period may be calculated by multiplying the label score of each content label of the previous time period of the current time period by an attenuation factor β (between 0 and 1, i.e., the first score attenuation parameter described above), thereby ensuring that the nearest label score is higher. Specifically, the expression (II) can be expressed as:
score day ＝score day-1 ×β，(0<β<1) The method comprises the steps of carrying out a first treatment on the surface of the Formula II
Wherein score day A tag score for the current time period; score day+1 A tag score for a time period preceding the current time period.
Thus, a tag score for each content tag after the current time period is obtained. And then, according to the label score of each content label after the current time period, selecting the content label with the label score meeting the score condition as a behavior label of the two-party behavior of the user. Of course, all content tags can also be used as behavior tags of the two-party behavior of the user. Thus, a behavior label set which is more in line with the behavior characteristics of the nearest two parties of the user and a label score of each behavior label are obtained.
(2) And (5) a multi-source fusion processing flow. In practical applications, a user is likely to have behaviors in multiple behavior sources, and meanwhile, the behavior labels of the user cannot be infinitely large, and TopN (for example, N is 20 or 50) is generally reserved. In this case, there are two methods of feature processing, one is to directly process multiple features, one for each behavior source, but the features are independent, so that it is difficult to explicitly establish the relationship between the features by using an upper model, and the complexity of the model is increased; another approach is to directly weight sum the behavior tags of multiple sources to process a feature, but this can overwhelm the real features of many users. For example, the user may behave against the tag "AAA" in every scene, possibly only once, but this is a key feature that needs to be preserved, and if the feature is processed in the above two ways, this behavior may be filtered out because of the frequency being too low. Thus, in embodiments of the present application, the feature processing of the multisource fusion is split into two steps, including:
1) Counting the occurrence number of each behavior label in all behavior sources, and preferentially reserving the behavior labels with the occurrence number exceeding a threshold value of the occurrence number (such as 80% of the total number). 2) And calculating tag scores of the rest behavior tags by using a multi-source weighted summation mode, wherein the weight of each behavior source can be customized, and comparability between the relative behavior sources is ensured. For example, the weights of the behavior sources are such that: search > read view using APP > (video/article) to forward comments, etc > (video/article). Thus, comprehensive multi-source behavior characteristics which are more in line with the reality of the user are obtained.
(3) And (5) a label filtering processing flow. Through the process flows of (1) and (2), each user has a plurality of behavior tags related to the two-way behavior. In actual practice, however, not every behavior label is associated with a promotional information recommendation goal (e.g., click through rate, conversion rate, etc.). For example, if the target time period is during spring festival, most users have content tags such as "spring festival", "past year", etc., so that the content tags cannot be distinguished by the user, the learning difficulty of the model for promoting information recommendation is increased, and the storage space is wasted, so that the behavior tags can be filtered.
In some embodiments, tag filtering may be implemented by: 1) And acquiring an interaction parameter value CTCVR (CTCVR is click rate, conversion rate and represents conversion probability of the user exposure popularization information) of the first object group aiming at the target popularization information. 2) And for each behavior label, defining a second object group with the behavior label, and then calculating a target interaction parameter value CTCVR of the second object group aiming at target popularization information. 3) And determining whether the target interaction parameter value is larger than the interaction parameter value, if so, indicating that the behavior label has the significance of user distinction, and if not, removing.
In some embodiments, tag filtering may also be implemented by: and scoring the combination of the < user, the behavior label > by adopting a pre-trained scoring model to obtain a scoring score, so that the label filtering is carried out according to the scoring score, and for example, the behavior labels with the scoring scores being ranked in descending order can be removed. This enables more refined tag filtering.
Finally, obtaining the filtered behavior label of each user, and then taking the filtered behavior label as the two-way behavior characteristic of the user. Of course, some behavior tags (e.g., the top N behavior tags in descending order of tag scores) may also be selected as the user's two-way behavior feature.
Second, a feature application section in the embodiment of the present application is explained. Because the two-party behavior feature and the behavior aiming at the popularization information are not in direct relation, in order to improve the effect of the two-party behavior feature on the recommendation system of the popularization information, the embodiment of the application provides the following two modes:
1) And carrying out characteristic cross processing on the two-party behavior characteristics and the popularization information characteristics. In some embodiments, the matching relationship between the two-party behavioral characteristics and the promotional information characteristics may be explicitly established through a feature intersection process. When the promotion information to be recommended is selected, the promotion information with the matched promotion information characteristics and the two-party behavior characteristics is preferentially selected, so that the effect of the two-party behavior characteristics on a recommendation system of the promotion information is improved. Here, the promotion information feature may refer to a promotion information tag, a classification category of promotion information, and the like. Thus, the model for promoting information recommendation can be learned: if the user has the two-party behavior feature A, clicking or converting the popularization information with the popularization information label of B is easier.
2) And modeling the independent subnetworks of the two-party behavior characteristics. In some embodiments, a sub-network (e.g., constructed by a neural network (e.g., convolutional neural network, deep neural network)) may be added separately to model the two-way behavioral features to splice with the non-two-way behavioral features of the user to arrive at the final features. In this way, the weight of the two-party behavior feature is increased by increasing the model parameters, so that the effect of the two-party behavior feature on the recommendation system of the popularization information is improved.
By applying the embodiment of the application, the two-party behavior of the user is subjected to feature processing, and the two-party behavior features of the user are introduced, so that the user is recommended to the popularization information through the two-party behavior features, the conversion of the popularization information can be promoted, and the popularization effect of the popularization information is improved.
Continuing with the description below of an exemplary architecture of feature acquisition device 555 implemented as a software module provided by embodiments of the present application, in some embodiments, as shown in fig. 2, the software modules stored in feature acquisition device 555 of memory 550 may include: the first obtaining module 5551 is configured to obtain interaction content corresponding to a target interaction behavior of a target object, where the target interaction behavior does not include interaction behavior for popularization information; the extraction module 5552 is configured to perform label extraction on the interactive content to obtain a plurality of behavior labels of the target object; a second obtaining module 5553, configured to obtain a tag score of each of the behavior tags; wherein the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, the behavior feature being used to recommend promotional information to the target object; a selecting module 5554, configured to select at least one target behavior label from the plurality of behavior labels based on the label score of each behavior label, and use the at least one target behavior label as a behavior feature of the target object.
In some embodiments, the extracting module 5552 is further configured to perform classification processing on the interactive content in multiple classification dimensions, so as to obtain classification categories of the interactive content in each classification dimension; and taking the classification category of the interactive content in each classification dimension as a plurality of behavior labels of the target object.
In some embodiments, the extracting module 5552 is further configured to obtain a plurality of candidate tags; respectively carrying out score prediction on each candidate tag to obtain a prediction score of each candidate tag, wherein the prediction score indicates the possibility degree that the candidate tag is a behavior tag of the target object; and taking the plurality of candidate labels with the predictive scores meeting the predictive score condition as a plurality of behavior labels of the target object.
In some embodiments, the second obtaining module 5553 is further configured to obtain a tag score of each of the behavior tags in the current time period; the second obtaining module 5553 is further configured to obtain a first fractional attenuation parameter, and for each behavior tag, perform the following processing respectively: when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and taking the initial tag score as the tag score of the behavior tag in the current time period; and when the current time period is the j-th time period for determining the tag score, determining the tag score of the behavior tag in the current time period based on the first score attenuation parameter and the tag score of the behavior tag in the (j-1) -th time period, wherein j is an integer greater than 1.
In some embodiments, the second obtaining module 5553 is further configured to obtain a tag score of each of the behavior tags in the current time period; the second obtaining module 5553 is further configured to obtain a second fractional attenuation parameter, and for each behavior tag, perform the following processing respectively: when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and determining the tag score of the behavior tag in the current time period based on the initial tag score and the behavior indication parameter of the current time period; determining a tag score of the behavior tag in a current time period based on the second score decay parameter, the tag score of the behavior tag in the (i-1) th time period, the initial tag score, and the behavior indication parameter when the current time period is the i-th time period in which the tag score is determined; the behavior indication parameter indicates whether a target interaction behavior corresponding to the behavior tag exists in the current time period, and i is an integer greater than 1.
In some embodiments, the second obtaining module 5553 is further configured to determine a relevance of the behavior tag to the interactive content; and determining an initial tag score of the behavior tag based on the correlation degree corresponding to the behavior tag, wherein the initial tag score and the correlation degree are in a positive correlation relationship.
In some embodiments, the selecting module 5554 is further configured to determine, for each of the behavior tags, a behavior source corresponding to the behavior tag, and determine the number of occurrences of the behavior tag in the behavior source; selecting first behavior tags with the number meeting a number condition from the behavior tags; determining a target tag score of a second behavior tag, which is other than the first behavior tag, according to the tag score of the second behavior tag and the behavior source weight of each behavior source corresponding to the second behavior tag, aiming at the second behavior tag in the plurality of behavior tags; selecting a third behavior label with the target label score meeting a score condition from the second behavior labels; and taking the first behavior label and the third behavior label as the at least one target behavior label.
In some embodiments, the selecting module 5554 is further configured to determine a behavior source corresponding to each behavior tag, and obtain a behavior source weight of each behavior source; determining a target tag score of each behavior tag based on the tag score of the behavior tag and the behavior source weight of each behavior source corresponding to the behavior tag; and selecting at least one fourth behavior label with the target label score meeting a score condition from the plurality of behavior labels, and taking the at least one fourth behavior label as the at least one target behavior label.
In some embodiments, when the number of the target behavior tags is a plurality, the selecting module 5554 is further configured to obtain, after the tag score based on each of the behavior tags selects at least one target behavior tag from the plurality of behavior tags, an interaction parameter value of a first object group for target promotion information, where the target promotion information carries the at least one target behavior tag, where the interaction parameter value indicates a likelihood that the object group performs an interaction for the target promotion information; for each target behavior label, determining a second object group with the target behavior label, and determining a target interaction parameter value of the second object group for the target popularization information; selecting a fifth behavior label with the target interaction parameter value larger than the interaction parameter value from a plurality of target behavior labels; the selecting module 5554 is further configured to use the fifth behavior label as a behavior feature of the target object.
In some embodiments, when the number of the target behavior tags is a plurality, the selecting module 5554 is further configured to, after selecting at least one target behavior tag from the plurality of behavior tags based on the tag score of each of the behavior tags, perform score prediction on each of the target behavior tags to obtain a prediction score, where the prediction score indicates a correlation between the target behavior tag and a recommended target of the promotion information; selecting a sixth behavior label with the predictive score meeting a score condition from a plurality of target behavior labels; the act of taking the at least one target act tag as the act feature of the target object includes: and taking the sixth behavior label as the behavior characteristic of the target object.
In some embodiments, the selecting module 5554 is further configured to obtain, after the at least one target behavior tag is used as a behavior feature of the target object, a matching relationship between the behavior feature and a promotion information feature, and obtain a promotion information feature to be recommended of each of a plurality of promotion information to be recommended; determining target popularization information characteristics matched with the behavior characteristics in the to-be-recommended popularization information characteristics based on the matching relation; and sending the popularization information to be recommended, which has the characteristics of the target popularization information, from the plurality of popularization information to be recommended to the terminal of the target object.
It should be noted that, the description of the embodiments of the device in this application is similar to the description of the embodiments of the method described above, and has similar beneficial effects as the embodiments of the method, which are not described herein. The technical details that are not described in the feature acquiring apparatus provided in the embodiment of the present application may be understood based on the description of the technical details in the foregoing method embodiment.
Embodiments of the present application also provide a computer program product comprising computer-executable instructions or a computer program stored in a computer-readable storage medium. The processor of the electronic device reads the computer-executable instructions or the computer program from the computer-readable storage medium, and the processor executes the computer-executable instructions or the computer program, so that the electronic device executes the feature acquisition method provided by the embodiment of the application.
The present embodiments also provide a computer-readable storage medium having stored therein computer-executable instructions or a computer program which, when executed by a processor, cause the processor to perform the feature acquisition method provided by the embodiments of the present application.
In some embodiments, the computer readable storage medium may be RAM, ROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; but may be a variety of devices including one or any combination of the above memories.
In some embodiments, computer-executable instructions may be written in any form of programming language, including compiled or interpreted languages, or declarative or procedural languages, in the form of programs, software modules, scripts, or code, and they may be deployed in any form, including as stand-alone programs or as modules, components, subroutines, or other units suitable for use in a computing environment.
As an example, computer-executable instructions may, but need not, correspond to files in a file system, may be stored as part of a file that holds other programs or data, such as in one or more scripts in a hypertext markup language (Hyper Text Markup Language, HTML) document, in a single file dedicated to the program in question, or in multiple coordinated files (e.g., files that store one or more modules, sub-programs, or portions of code).
As an example, computer-executable instructions may be deployed to be executed on one electronic device or on multiple electronic devices located at one site or, alternatively, on multiple electronic devices distributed across multiple sites and interconnected by a communication network.
The foregoing is merely exemplary embodiments of the present application and is not intended to limit the scope of the present application. Any modifications, equivalent substitutions, improvements, etc. that are within the spirit and scope of the present application are intended to be included within the scope of the present application.

Claims (15)

1. A method of feature acquisition, the method comprising:
acquiring interactive content corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at popularization information;
extracting tags from the interactive content to obtain a plurality of behavior tags of the target object;
acquiring label scores of the behavior labels;
wherein the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, the behavior feature being used to recommend promotional information to the target object;
and selecting at least one target behavior label from the plurality of behavior labels based on the label score of each behavior label, and taking the at least one target behavior label as the behavior characteristic of the target object.
2. The method of claim 1, wherein the extracting the tags from the interactive content to obtain the plurality of behavior tags for the target object comprises:
performing classification processing of a plurality of classification dimensions on the interactive content to obtain classification categories of the interactive content in each classification dimension;
and taking the classification category of the interactive content in each classification dimension as a plurality of behavior labels of the target object.
3. The method of claim 1, wherein the extracting the tags from the interactive content to obtain the plurality of behavior tags for the target object comprises:
acquiring a plurality of candidate labels;
respectively carrying out score prediction on each candidate tag to obtain a prediction score of each candidate tag, wherein the prediction score indicates the possibility degree that the candidate tag is a behavior tag of the target object;
and taking the plurality of candidate labels with the predictive scores meeting the predictive score condition as a plurality of behavior labels of the target object.
4. The method of claim 1, wherein the obtaining the tag score for each of the behavior tags comprises: acquiring label scores of the behavior labels in the current time period;
The step of obtaining the tag score of each behavior tag in the current time period comprises the following steps:
acquiring a first fractional attenuation parameter, and for each behavior label, respectively executing the following processing:
when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and taking the initial tag score as the tag score of the behavior tag in the current time period;
and when the current time period is the j-th time period for determining the tag score, determining the tag score of the behavior tag in the current time period based on the first score attenuation parameter and the tag score of the behavior tag in the (j-1) -th time period, wherein j is an integer greater than 1.
5. The method of claim 1, wherein the obtaining the tag score for each of the behavior tags comprises: acquiring label scores of the behavior labels in the current time period;
the step of obtaining the tag score of each behavior tag in the current time period comprises the following steps:
acquiring a second fractional attenuation parameter, and for each behavior label, respectively executing the following processing:
when the current time period is the 1 st time period for determining the tag score, acquiring an initial tag score of the behavior tag, and determining the tag score of the behavior tag in the current time period based on the initial tag score and the behavior indication parameter of the current time period;
Determining a tag score of the behavior tag in a current time period based on the second score decay parameter, the tag score of the behavior tag in the (i-1) th time period, the initial tag score, and the behavior indication parameter when the current time period is the i-th time period in which the tag score is determined;
the behavior indication parameter indicates whether a target interaction behavior corresponding to the behavior tag exists in the current time period, and i is an integer greater than 1.
6. The method of claim 4 or 5, wherein the obtaining the initial tag score for the behavioral tag comprises:
determining the relativity of the behavior label and the interactive content;
and determining an initial tag score of the behavior tag based on the correlation degree corresponding to the behavior tag, wherein the initial tag score and the correlation degree are in a positive correlation relationship.
7. The method of claim 1, wherein selecting at least one target behavioral label from the plurality of behavioral labels based on the label score for each of the behavioral labels comprises:
for each behavior label, determining a behavior source corresponding to the behavior label, and determining the occurrence number of the behavior label in the behavior source;
Selecting first behavior tags with the number meeting a number condition from the behavior tags;
determining a target tag score of a second behavior tag, which is other than the first behavior tag, according to the tag score of the second behavior tag and the behavior source weight of each behavior source corresponding to the second behavior tag, aiming at the second behavior tag in the plurality of behavior tags;
selecting a third behavior label with the target label score meeting a score condition from the second behavior labels;
and taking the first behavior label and the third behavior label as the at least one target behavior label.
8. The method of claim 1, wherein selecting at least one target behavioral label from the plurality of behavioral labels based on the label score for each of the behavioral labels comprises:
determining a behavior source corresponding to each behavior label, and acquiring a behavior source weight of each behavior source;
determining a target tag score of each behavior tag based on the tag score of the behavior tag and the behavior source weight of each behavior source corresponding to the behavior tag;
And selecting at least one fourth behavior label with the target label score meeting a score condition from the plurality of behavior labels, and taking the at least one fourth behavior label as the at least one target behavior label.
9. The method of claim 1, wherein when the number of target behavior tags is a plurality, the method further comprises, after selecting at least one target behavior tag from the plurality of behavior tags based on the tag score of each of the behavior tags:
acquiring an interaction parameter value of a first object group aiming at target popularization information, wherein the target popularization information carries the at least one target behavior label, and the interaction parameter value indicates the possibility of the object group to execute interaction behaviors aiming at the target popularization information;
for each target behavior label, determining a second object group with the target behavior label, and determining a target interaction parameter value of the second object group for the target popularization information;
selecting a fifth behavior label with the target interaction parameter value larger than the interaction parameter value from a plurality of target behavior labels;
the act of taking the at least one target act tag as the act feature of the target object includes: and taking the fifth behavior label as the behavior characteristic of the target object.
10. The method of claim 1, wherein when the number of target behavior tags is a plurality, the method further comprises, after selecting at least one target behavior tag from the plurality of behavior tags based on the tag score of each of the behavior tags:
aiming at each target behavior label, carrying out score prediction on the target behavior label to obtain a prediction score, wherein the prediction score indicates the correlation between the target behavior label and a recommended target of popularization information;
selecting a sixth behavior label with the predictive score meeting a score condition from a plurality of target behavior labels;
the act of taking the at least one target act tag as the act feature of the target object includes: and taking the sixth behavior label as the behavior characteristic of the target object.
11. The method of claim 1, wherein after said characterizing the at least one target behavior tag as a behavior of the target object, the method further comprises:
acquiring a matching relation between the behavior characteristic and the popularization information characteristic, and acquiring the popularization information characteristic to be recommended of each of a plurality of popularization information to be recommended;
Determining target popularization information characteristics matched with the behavior characteristics in the to-be-recommended popularization information characteristics based on the matching relation;
and sending the popularization information to be recommended, which has the characteristics of the target popularization information, from the plurality of popularization information to be recommended to the terminal of the target object.
12. A feature acquisition apparatus, the apparatus comprising:
the first acquisition module is used for acquiring interactive contents corresponding to target interactive behaviors of a target object, wherein the target interactive behaviors do not comprise interactive behaviors aiming at popularization information;
the extraction module is used for extracting the tags of the interactive contents to obtain a plurality of behavior tags of the target object;
the second acquisition module is used for acquiring the label score of each behavior label;
wherein the tag score indicates a likelihood that the behavior tag is a behavior feature of the target object, the behavior feature being used to recommend promotional information to the target object;
the selecting module is used for selecting at least one target behavior label from the plurality of behavior labels based on the label score of each behavior label, and taking the at least one target behavior label as the behavior characteristic of the target object.
13. An electronic device, the electronic device comprising:
a memory for storing computer executable instructions;
a processor for implementing the feature acquisition method of any one of claims 1 to 11 when executing computer-executable instructions stored in the memory.
14. A computer-readable storage medium storing computer-executable instructions or a computer program, which, when executed by a processor, implements the feature acquisition method of any one of claims 1 to 11.
15. A computer program product comprising computer executable instructions or a computer program which, when executed by a processor, implements the feature acquisition method of any one of claims 1 to 11.
CN202311326921.2A 2023-10-12 2023-10-12 Feature acquisition method, device, apparatus, storage medium, and program product Pending CN117390577A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311326921.2A CN117390577A (en) 2023-10-12 2023-10-12 Feature acquisition method, device, apparatus, storage medium, and program product

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311326921.2A CN117390577A (en) 2023-10-12 2023-10-12 Feature acquisition method, device, apparatus, storage medium, and program product

Publications (1)

Publication Number Publication Date
CN117390577A true CN117390577A (en) 2024-01-12

Family

ID=89436785

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311326921.2A Pending CN117390577A (en) 2023-10-12 2023-10-12 Feature acquisition method, device, apparatus, storage medium, and program product

Country Status (1)

Country Link
CN (1) CN117390577A (en)

Similar Documents

Publication Publication Date Title
US8156138B2 (en) System and method for providing targeted content
CN111444428A (en) Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium
CN1934569B (en) Search systems and methods with integration of user annotations
US9639846B2 (en) System and method for providing targeted content
WO2022016522A1 (en) Recommendation model training method and apparatus, recommendation method and apparatus, and computer-readable medium
US20230162005A1 (en) Neural network distillation method and apparatus
CN110825956A (en) Information flow recommendation method and device, computer equipment and storage medium
US8560606B2 (en) Social network informed mashup creation
CN108121814B (en) Search result ranking model generation method and device
CN110737811B (en) Application classification method and device and related equipment
CN111259220B (en) Data acquisition method and system based on big data
WO2013070534A1 (en) Function extension for browsers or documents
CN110110233A (en) Information processing method, device, medium and calculating equipment
CN106471497A (en) Auxiliary using context browses
CN112819024B (en) Model processing method, user data processing method and device and computer equipment
CN114036398A (en) Content recommendation and ranking model training method, device, equipment and storage medium
CN112328889A (en) Method and device for determining recommended search terms, readable medium and electronic equipment
CN114428910A (en) Resource recommendation method and device, electronic equipment, product and medium
US10387934B1 (en) Method medium and system for category prediction for a changed shopping mission
CN116823410A (en) Data processing method, object processing method, recommending method and computing device
CN113641916B (en) Content recommendation method and device, electronic equipment and storage medium
CN116204709A (en) Data processing method and related device
JP2024507029A (en) Web page identification methods, devices, electronic devices, media and computer programs
CN117390577A (en) Feature acquisition method, device, apparatus, storage medium, and program product
CN116484085A (en) Information delivery method, device, equipment, storage medium and program product

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication