CN115544358A - Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics - Google Patents

Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics Download PDF

Info

Publication number
CN115544358A
CN115544358A CN202211186906.8A CN202211186906A CN115544358A CN 115544358 A CN115544358 A CN 115544358A CN 202211186906 A CN202211186906 A CN 202211186906A CN 115544358 A CN115544358 A CN 115544358A
Authority
CN
China
Prior art keywords
task
module
user
live broadcast
live
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
CN202211186906.8A
Other languages
Chinese (zh)
Inventor
崔永庆
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.)
Quantuo Technology Hangzhou Co ltd
Original Assignee
Quantuo Technology Hangzhou 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 Quantuo Technology Hangzhou Co ltd filed Critical Quantuo Technology Hangzhou Co ltd
Priority to CN202211186906.8A priority Critical patent/CN115544358A/en
Publication of CN115544358A publication Critical patent/CN115544358A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations

Landscapes

  • Engineering & Computer Science (AREA)
  • Databases & Information Systems (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Physics & Mathematics (AREA)
  • Economics (AREA)
  • General Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Marketing (AREA)
  • Development Economics (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a short video live broadcasting marketing recommendation system and method fusing multi-platform behavior characteristics, and the system comprises a data acquisition module, an acquired data preprocessing module, a recommendation system historical behavior data preprocessing module, a live broadcasting task recall establishing model module, a live broadcasting task database storage module, a live broadcasting task sequencing establishing model module and a live broadcasting task recommendation module, wherein the data acquisition module is in control connection with the acquired data preprocessing module, the acquired data preprocessing module and the recommendation system historical behavior data preprocessing module are in control connection with the live broadcasting task recall establishing model module, and the live broadcasting task recall establishing model module is in control connection with the live broadcasting task database storage module; the method and the system make up the defect that a short video live broadcasting marketing system is lack of a recommendation function, provide a task recommendation function for broadcasting main user selections, assist the main users in quickly selecting live broadcasting tasks, and provide an efficient task selection scheme for short video live broadcasting marketing.

Description

Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics
Technical Field
The invention relates to the technical field of recommendation systems, in particular to a short video live broadcast marketing recommendation system and method integrating multi-platform behavior characteristics.
Background
Today, with the rapid development of network technologies, short videos attract more and more people's attention by virtue of the characteristics of low threshold, convenient use and strong social participation. The short video live broadcast marketing is a product combining electronic commerce and short video live broadcast, live broadcast with goods is carried out in a short video live broadcast mode, and the rapid change of broadcast main flow is promoted by utilizing the huge user base number and the convenient live broadcast platform of the short video platform. The short video live broadcast marketing can well make up for short boards with insufficient commodity description and high price in the traditional e-commerce platform, so that a transaction link between brand merchants and consumers is constructed, and the short video live broadcast marketing has important significance for optimizing an industrial structure and realizing the coordinated development of the industry.
However, in the existing live broadcast marketing recommendation system, recommendation is mainly performed according to the historical behavior record of the user, the purchasing characteristics of the consumer are lacked, and the marketing task is difficult to be accurately recommended, so that a certain limit exists between a broadcaster and a brand owner, and the broadcaster cannot select products according to the own shipment style. Therefore, a short video live broadcast marketing recommendation system for collecting multi-platform user characteristics is urgently needed, the commodity data of a user shop and the historical behavior sequence of the user are fused, a credible third party recommendation and transaction platform is provided for a broadcaster and a brand merchant, and electronic commerce selection is assisted.
Disclosure of Invention
The invention aims to provide a short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics, and aims to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: a short video live broadcast marketing recommendation method integrating multi-platform behavior characteristics comprises the following steps: step one, collecting; step two, collected data are preprocessed; step three, preprocessing historical behavior data; step four, establishing a model; step five, storing; step six, predicting; step seven, recommending;
in the first step, user data of a plurality of platforms are acquired through a set data acquisition module;
in the second step, the collected data is preprocessed and persisted by using a collected data preprocessing module, and the preprocessing of the collected data comprises the preprocessing of commodity data of a shop of a user and demographic information of the user;
in the third step, a recommendation system historical behavior data preprocessing module is used for preprocessing the recommendation system historical behavior data;
in the fourth step, a live broadcast task recall model is established by the live broadcast task recall model establishing module, and then the preprocessed user store commodity data, the preprocessed demographic information and the historical behavior data of the recommendation system are used as the input of the live broadcast task recall model to obtain a live broadcast task sequence to be recommended;
in the fifth step, the live broadcast task sequence to be recommended is stored in the database in a persistent mode through a live broadcast task database storage module;
in the sixth step, the establishment of a live broadcast task ordering model is completed by utilizing a live broadcast task ordering model establishment module, the live broadcast task ordering model takes the preprocessed user shop commodity sequence and each live broadcast task to be recommended as input, and the probability of clicking the live broadcast task by the user is predicted;
and in the seventh step, the set live task recommendation module is used, and when the broadcasting master user logs in the live marketing recommendation system, the task with the highest click rate is taken out from the live task recommendation database and recommended to the broadcasting master user.
Preferably, in the step one, the data acquisition module includes:
1) Collecting user shop commodity characteristics and collected user shop commodity data by using Scapy crawler frame
Figure BDA0003867871000000021
Wherein x is s The data of the commercial products of the shop on behalf of the user,
Figure BDA0003867871000000022
indicates the kth user store commodity data,
Figure BDA0003867871000000031
the price of the goods is expressed and,
Figure BDA0003867871000000032
which indicates the time at which the goods are released,
Figure BDA0003867871000000033
which is indicative of the type of the article,
Figure BDA0003867871000000034
indicating commodity sales;
2) Collects the K recently issued by the user r Individual item information defining a user store item sequence
Figure BDA0003867871000000035
3) Collecting demographic information X of broadcasting main users P ＝[p sex ,p age ,p position ] T ，p sex Representing the gender, p, of the user age Indicates the age, p, of the user position Representing geographical location information.
Preferably, in the second step, the collected data preprocessing module specifically works as follows:
1) The pre-processing of the store commodity data comprises the processing of commodity price, commodity release time, commodity type and commodity sales volume, and the user store commodity data is preprocessed into
Figure BDA0003867871000000036
The pre-processed user shop commodity sequence is
Figure BDA0003867871000000037
2) The preprocessing of the acquired data also includes preprocessing demographic information of the broadcasting master user to obtain processed demographic characteristics
Figure BDA0003867871000000038
3) The latest K of the processed user shop commodity is processed by using a MySQL relational database in the persistence processing of the acquired data r A sequence
Figure BDA0003867871000000039
And demographic information
Figure BDA00038678710000000310
And performing persistent storage.
Preferably, in the third step, the recommendation system historical behavior data preprocessing module specifically works as follows:
1) Defining a single historical behavioral data of a live marketing recommendation system as
Figure BDA00038678710000000311
Wherein x is b A single historical behavioral data representing the broadcasting master user,
Figure BDA00038678710000000312
denotes the kth b The historical data of the behavior of the user is stored,
Figure BDA00038678710000000313
representing a task budget, representing
Figure BDA00038678710000000314
The time at which the task was issued,
Figure BDA00038678710000000315
indicating that the user's browsing duration is played,
Figure BDA00038678710000000316
the status of the task is represented and,
Figure BDA00038678710000000317
indicating the number of the task-associated loaded commodities;
2) Selecting the nearest K of the user r Historical behavioral data, deficiency of K r The part of the user data is filled by using a default value to obtain the latest K of the broadcast main user r A historical behavior data sequence K b Is defined as
Figure BDA0003867871000000041
3) Preprocessing broadcast master user recent K r A historical behavior data sequence X b The single historical behavior characteristic after the preprocessing is obtained as
Figure BDA0003867871000000042
Historical behavior sequencesIs composed of
Figure BDA0003867871000000043
K for processed historical behavioral data sequence using MySQL relational database r A sequence
Figure BDA0003867871000000044
Performing persistent storage of sequences
Figure BDA0003867871000000045
And performing persistent storage.
Preferably, in the fourth step, the specific work of establishing the live task recall model module is as follows:
1) Defining live tasks as
Figure BDA0003867871000000046
t budget For task budgeting, t pos For task release time, t state Is the task state, t telate Associating the number of items in stock for the task, t model A task mode is adopted;
2) Input of live task recall model is user store commodity sequence after preprocessing
Figure BDA0003867871000000047
Demographic information
Figure BDA0003867871000000048
And historical behavior sequence of live marketing recommendation system
Figure BDA0003867871000000049
Outputting a sequence of central tasks represented by a marketing task type of interest to a user
Figure BDA00038678710000000410
3) The direct broadcast task recall model calculates the middle of the marketing task type represented by the marketing task type by using a k nearest neighbor methodClosest N in the heart task f Individual tasks forming a set of recalled tasks
Figure BDA00038678710000000411
N h The number of tasks recalled.
Preferably, in the fifth step, the specific work of the live task database storage module is to collect the task set T obtained in the fourth step h And persisting to the live task database.
Preferably, in the sixth step, the specific work of establishing the live task sequencing model module is as follows:
1) The input of the live broadcast task sequencing model is a user shop commodity sequence after preprocessing
Figure BDA0003867871000000051
And recall tasks
Figure BDA0003867871000000052
2) The live broadcast task sequencing model obtains a predicted click rate, and the tasks are sequenced from large to small according to the click rate to obtain a final recommended task sequence
Figure BDA0003867871000000053
Preferably, in the seventh step, after the live main user logs in the live marketing recommendation system, the N with the highest click rate is taken out from the live task recommendation database top And recommending the tasks to the broadcasting master user as final marketing recommendation tasks and recommending the tasks to the user on the page.
A short video live broadcasting marketing recommendation system fusing multi-platform behavior characteristics comprises a data acquisition module, a collected data preprocessing module, a recommendation system historical behavior data preprocessing module, a live broadcasting task recall model establishing module, a live broadcasting task database storage module, a live broadcasting task sequencing model establishing module and a live broadcasting task recommendation module, wherein the data acquisition module is in control connection with the collected data preprocessing module, the collected data preprocessing module and the recommendation system historical behavior data preprocessing module are in control connection with the live broadcasting task recall model establishing module, the live broadcasting task recall model establishing module is in control connection with the live broadcasting task database storage module, the live broadcasting task database storage module is in control connection with the live broadcasting task sequencing model establishing module, and the live broadcasting task sequencing model establishing module is in control connection with the live broadcasting task recommendation module.
Compared with the prior art, the invention has the beneficial effects that: the short-video live broadcasting marketing system overcomes the defect of lack of a recommendation function in the short-video live broadcasting marketing system, provides a task recommendation function for broadcasting the selection of the main user, facilitates the broadcasting of the task which is selected by the main user and accords with the own live broadcasting style, improves the user experience of the system, and enables the short-video live broadcasting marketing to increase.
Drawings
FIG. 1 is a schematic diagram of the operation of the present invention;
FIG. 2 is a flow chart of the method of the present invention;
FIG. 3 is a schematic view of the overall module structure of the present invention;
in the figure: 1. a data acquisition module; 2. a collected data preprocessing module; 3. a recommendation system historical behavior data preprocessing module; 4. establishing a live broadcast task recall model module; 5. a live broadcast task database storage module; 6. establishing a live broadcast task sequencing model module; 7. and a live broadcast task recommendation module.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Referring to fig. 1-2, an embodiment of the present invention is shown: a short video live broadcast marketing recommendation method integrating multi-platform behavior characteristics comprises the following steps: step one, collecting; step two, collected data are preprocessed; step three, preprocessing historical behavior data; step four, establishing a model; step five, storing; step six, predicting; step seven, recommending;
in the first step, user data of a plurality of platforms are acquired through the set data acquisition module 1; the data acquisition module 1 mainly uses a script crawler frame to acquire the store commodity characteristics of the user, and the acquired user store commodity data is defined as follows:
Figure BDA0003867871000000061
wherein x is s The data of the commercial products of the shop on behalf of the user,
Figure BDA0003867871000000062
indicates the kth user store commodity data,
Figure BDA0003867871000000063
the price of the goods is expressed and,
Figure BDA0003867871000000064
which indicates the time at which the goods are released,
Figure BDA0003867871000000065
which is indicative of the type of the article,
Figure BDA0003867871000000066
indicating commodity sales; collects the K recently issued by the user r The individual item information, which defines the user store item sequence, is defined as follows:
Figure BDA0003867871000000067
gathering demographic information for the broadcasting user is defined as follows:
X P ＝[p sex ,p age ,p position ] T
wherein p is sex Representing the gender, p, of the user age Indicates the age, p, of the user position To representGeographic location information;
in the second step, the preprocessing of the store commodity data in the collected data preprocessing comprises the processing of commodity price, commodity release time, commodity type and commodity sales volume, and the processing of the kth commodity data
Figure BDA0003867871000000071
For its commodity price
Figure BDA0003867871000000072
The normalization method is adopted, and the treatment process is as follows:
Figure BDA0003867871000000073
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003867871000000074
for the normalized commodity price, e max Maximum number of prices of goods, e min Is the minimum number of commodity prices; commodity release time in data acquisition preprocessing
Figure BDA0003867871000000075
The preprocessing of (1) converting the commodity release time into a time stamp
Figure BDA0003867871000000076
Processing difference value with current time stamp to obtain time difference
Figure BDA0003867871000000077
The treatment process is as follows:
Figure BDA0003867871000000078
wherein, T c Is the current timestamp; normalizing the time difference by using an arc tangent function to calculate the commodity release time after preprocessing
Figure BDA0003867871000000079
The process is as follows:
Figure BDA00038678710000000710
for commodity type in data collection preprocessing
Figure BDA00038678710000000711
The data preprocessing part adopts one-hot coding to expand discrete commodity type characteristics to European space to obtain the processed commodity type
Figure BDA00038678710000000712
Sales volume for goods in data acquisition pre-processing
Figure BDA0003867871000000081
The preprocessing of (3) adopts z-score standardization processing, and firstly calculates the type of the commodity to which the commodity belongs
Figure BDA0003867871000000082
The standard deviation of commodity sales of (1) is calculated as follows:
Figure BDA0003867871000000083
wherein N is t The number of the same type goods under the type of the goods,
Figure BDA0003867871000000084
for the sales volume of the ith commodity, mu represents the mean value of the sales volumes of the commodities under the unified type, and sigma represents the calculated standard deviation; sales volume of commodities
Figure BDA0003867871000000085
Performing z-score standardization, and calculating commodity sales after pretreatment
Figure BDA0003867871000000086
Figure BDA0003867871000000087
The user store commodity data after being preprocessed are as follows:
Figure BDA0003867871000000088
user store commodity nearest K r The sequences are:
Figure BDA0003867871000000089
preprocessing demographic information of the broadcasting main user, and broadcasting gender P of the main user sex And geographical location information P position One-hot was used to obtain the processed gender characteristics
Figure BDA00038678710000000810
And geographic location information features
Figure BDA00038678710000000811
For user age P age The normalization process is performed in the following manner:
Figure BDA00038678710000000812
wherein the content of the first and second substances,
Figure BDA00038678710000000813
is the minimum value of the age of the patient,
Figure BDA00038678710000000814
is the maximum value of age;
people who broadcast main users after data preprocessingOral statistics information
Figure BDA0003867871000000091
Is composed of
Figure BDA0003867871000000092
The latest K of the processed user shop commodity is compared with the MySQL relational database in the persistence operation of the collected data r A sequence
Figure BDA0003867871000000093
And demographic information
Figure BDA0003867871000000094
Carrying out persistent storage;
in the third step, the historical behavior data of the recommendation system is preprocessed by using a historical behavior data preprocessing module 3 of the recommendation system; preprocessing historical behavior data of a recommendation system; the recommendation system historical behavior data comprises historical behavior data generated by a broadcasting master user when the broadcasting master user browses the short video live broadcast marketing recommendation system, and the single historical behavior data is defined as follows:
Figure BDA0003867871000000095
wherein x is b A single historical behavior data representing the broadcasting of the primary user,
Figure BDA0003867871000000096
denotes the kth b The historical behavior data of the user is stored in a memory,
Figure BDA0003867871000000097
representing a task budget, representing
Figure BDA0003867871000000098
The time at which the task was issued,
Figure BDA0003867871000000099
indicating that the browsing time of the main user is played,
Figure BDA00038678710000000910
the status of the task is represented and,
Figure BDA00038678710000000911
indicating the number of the task-associated loaded commodities; 2. selecting the nearest K of the user r Historical behavioral data, deficiency of K r The part of the user data is filled by using a default value to obtain the latest K of the broadcast main user r A historical behavior data sequence K b The definition is as follows:
Figure BDA00038678710000000912
budget for tasks
Figure BDA00038678710000000913
The pretreatment of (2) adopts a normalization method, and the treatment process comprises the following steps:
Figure BDA00038678710000000914
wherein the content of the first and second substances,
Figure BDA00038678710000000915
for the purpose of normalizing the task budget after normalization,
Figure BDA00038678710000000916
for the purpose of the minimum value of the budget,
Figure BDA00038678710000000917
is the maximum value of the budget;
time to task release in user historical behavior data
Figure BDA0003867871000000101
The preprocessing of (1) converting task release time into time stamp
Figure BDA0003867871000000102
Processing the difference value with the current timestamp to obtain the time difference
Figure BDA0003867871000000103
The treatment process is as follows:
Figure BDA0003867871000000104
wherein, T c Is the current timestamp; normalizing the processing time difference by using an arc tangent function, and calculating the task release time after preprocessing
Figure BDA0003867871000000105
The process is as follows:
Figure BDA0003867871000000106
browsing duration of on-air main user in user historical behavior data
Figure BDA0003867871000000107
The preprocessing of (2) first converts the browsing duration into a timestamp
Figure BDA0003867871000000108
Calculating and processing browsing duration by using arc tangent normalization function
Figure BDA0003867871000000109
The treatment process is as follows:
Figure BDA00038678710000001010
task state in user historical behavior data
Figure BDA00038678710000001011
The one-hot coding is adopted for preprocessing, and the processed task state is obtained
Figure BDA00038678710000001012
Associating number of goods in shipment to task in user historical behavior data
Figure BDA00038678710000001013
Preprocessing, namely using a normalization method, and comprising the following specific steps of:
Figure BDA00038678710000001014
wherein the content of the first and second substances,
Figure BDA00038678710000001015
the number of items in stock is associated for the task after normalization,
Figure BDA00038678710000001016
associating a minimum number of shipped items for the task,
Figure BDA0003867871000000111
associating the maximum number of the goods with the task; after pretreatment, the expression K b Individual historical behavioral data
Figure BDA0003867871000000112
Is shown as
Figure BDA0003867871000000113
Historical behavioral data sequence
Figure BDA0003867871000000114
Is composed of
Figure BDA0003867871000000115
K for recommending historical behavior data sequence completed by processing by using MySQL relational database in system historical behavior data preprocessing operation r A sequence
Figure BDA0003867871000000116
Performing persistent storage;
in the fourth step, the live broadcast task recall model establishing module 4 establishes a live broadcast task recall model, then preprocessed user store commodity data, demographic information and recommendation system historical behavior data are used as input of the live broadcast task recall model, the live broadcast task recall is used for recalling live broadcast tasks existing in the database, and the live broadcast tasks in the database are defined as follows:
Figure BDA0003867871000000117
wherein, t budget For task budgeting, t pos For task release time, t state For the task state, t telate Associating the number of loaded items for the task, t model A task mode is adopted; the live broadcast task recall model is a simple neural network classification model, and the input of the live broadcast task recall model is a preprocessed user shop commodity sequence
Figure BDA0003867871000000118
Demographic information
Figure BDA0003867871000000119
And historical behavior sequence of live marketing recommendation system
Figure BDA00038678710000001110
Live broadcast task T in database l Is divided into N b A class, N b ≤N l Selecting the center live broadcasting marketing task represented by each category to form a set T d Expressed as:
Figure BDA00038678710000001111
wherein the content of the first and second substances,
Figure BDA00038678710000001112
for the central live marketing task set represented by each class, the output of the live task recall model is whether the live task recall model belongs to the set T d The task of (1), expressed as:
Figure BDA0003867871000000121
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA0003867871000000122
represents whether the user is on the kth d Individual task types are of interest; in the live broadcast task recall model, the preprocessed user shop commodity sequence is recalled
Figure BDA0003867871000000123
Demographic information
Figure BDA0003867871000000124
And historical behavior sequence of live marketing recommendation system
Figure BDA0003867871000000125
As input, the final model output is obtained
Figure BDA0003867871000000126
Get T d And
Figure BDA0003867871000000127
of intersection of
Figure BDA0003867871000000128
Is recorded as:
Figure BDA0003867871000000129
wherein, the first and the second end of the pipe are connected with each other,
Figure BDA00038678710000001210
center task represented by marketing task type of interest to user, N e For broadcasting ownerThe number of central tasks represented by the marketing task types in which the user is interested; the live broadcast task recall model calculates the most similar N in the central tasks represented by the marketing task types by using a k nearest neighbor method f The calculation process of the task is as follows:
Figure BDA00038678710000001211
wherein the content of the first and second substances,
Figure BDA00038678710000001212
x q ∈T d ，
Figure BDA00038678710000001213
composing a set of recalled tasks
Figure BDA00038678710000001214
Wherein N is h ≤(N e *N f )，N h Number of tasks recalled;
in the fifth step, the live broadcast task sequence to be recommended is stored in the database in a persistent mode through the live broadcast task database storage module 5; the concrete work of the live broadcast task database storage module 5 is to collect the task set T obtained in the step four h Persisting to a live task database;
in the sixth step, the establishment of a live broadcast task sequencing model is completed by utilizing the establishment of the live broadcast task sequencing model module 6, the live broadcast task sequencing model is a simple regression model, and the input of the live broadcast task sequencing model is a preprocessed user shop commodity sequence
Figure BDA0003867871000000131
And recall tasks
Figure BDA0003867871000000132
Output as
Probability P of user clicking the task click The input of the live task ranking model may be expressed as:
Figure BDA0003867871000000133
the input is collected as
Figure BDA0003867871000000134
The predicted click rate is obtained by broadcasting a task sequencing model, and the tasks are sequenced from large to small according to the click rate to obtain a final recommended task sequence
Figure BDA0003867871000000135
Inputting a set input live broadcast task sequencing model to obtain a predicted click rate as follows:
Figure BDA0003867871000000136
wherein the content of the first and second substances,
Figure BDA0003867871000000137
denotes the kth p The click probability of each task; the tasks are sorted from large to small according to the click rate to obtain a final recommended task sequence T rec Expressed as:
Figure BDA0003867871000000138
will T rec Persistently storing the data to a live task recommendation database;
in the seventh step, the set live broadcast task recommendation module 7 is used, and after the live broadcast main user logs in the live broadcast marketing recommendation system, the N with the highest click rate is taken out of the live broadcast task recommendation database top The tasks are recommended to the broadcasting master user, and the broadcasting master user can conveniently select the live marketing tasks meeting the interests of the broadcasting master user.
Referring to fig. 3, an embodiment of the present invention: a short video live broadcasting marketing recommendation system fusing multi-platform behavior characteristics comprises a data acquisition module 1, a data acquisition preprocessing module 2, a recommendation system historical behavior data preprocessing module 3, a live broadcasting task recall establishing model module 4, a live broadcasting task database storage module 5, a live broadcasting task sequencing establishing model module 6 and a live broadcasting task recommendation module 7, wherein the data acquisition module 1 is in control connection with the data acquisition preprocessing module 2, the data acquisition preprocessing module 2 and the recommendation system historical behavior data preprocessing module 3 are in control connection with a live broadcasting task recall establishing model module 4, the live broadcasting task recall establishing model module 4 is in control connection with the live broadcasting task database storage module 5, the live broadcasting task database storage module 5 is in control connection with a live broadcasting task sequencing establishing model module 6, and the live broadcasting task sequencing model module 6 is in control connection with a task recommendation module 7.
Based on the above, the invention has the advantages that: the invention can provide a task recommendation function for the broadcast main user selection, and is convenient for the broadcast main user to select the task according with the own live broadcast style, thereby improving the user experience of the system and enabling the short video live broadcast marketing growth.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.

Claims (9)

1. A short video live broadcast marketing recommendation method integrating multi-platform behavior characteristics comprises the following steps: step one, collecting; preprocessing collected data; step three, preprocessing historical behavior data; step four, establishing a model; step five, storing; step six, predicting; step seven, recommending; the method is characterized in that:
in the first step, user data of a plurality of platforms are acquired through a set data acquisition module (1);
in the second step, the collected data is preprocessed and persisted by using a collected data preprocessing module (2), and the preprocessing of the collected data comprises the preprocessing of commodity data of a shop of a user and demographic information of the user;
in the third step, a recommendation system historical behavior data preprocessing module (3) is used for preprocessing the recommendation system historical behavior data;
in the fourth step, a live broadcast task recall model is established through the live broadcast task recall model establishing module (4), and then preprocessed user store commodity data, demographic information and recommendation system historical behavior data are used as input of the live broadcast task recall model to obtain a live broadcast task sequence to be recommended;
in the fifth step, the live broadcast task sequence to be recommended is stored into the database in a persistent mode through the live broadcast task database storage module (5);
in the sixth step, the establishment of a live broadcast task sequencing model is completed by utilizing a live broadcast task sequencing model establishment module (6), the live broadcast task sequencing model takes the preprocessed user shop commodity sequence and each live broadcast task to be recommended as input, and the probability of clicking the live broadcast task by the user is predicted;
and in the seventh step, by using the set live task recommendation module (7), after the broadcasting main user logs in the live marketing recommendation system, the task with the highest click rate is taken out from the live task recommendation database and recommended to the broadcasting main user.
2. The short-video live marketing recommendation method integrating multi-platform behavior features according to claim 1, characterized by comprising the following steps of: in the first step, the data acquisition module (1) comprises the following working steps:
1) Collecting user shop commodity characteristics and collected user shop commodity data by using Scapy crawler frame
Figure FDA0003867870990000021
Wherein x is s The data of the commercial products of the shop on behalf of the user,
Figure FDA0003867870990000022
indicating the kth user store commodity data,
Figure FDA0003867870990000023
the price of the goods is represented and,
Figure FDA0003867870990000024
which indicates the time at which the goods are released,
Figure FDA0003867870990000025
which is indicative of the type of the article,
Figure FDA0003867870990000026
indicating commodity sales;
2) Collects the K recently issued by the user r Individual item information defining a user store item sequence
Figure FDA0003867870990000027
3) Collecting demographic information X of broadcasting main users P ＝[p sex ,p age ,p position ] T ，p sex Representing the gender, p, of the user age Indicates the age, p, of the user position Representing geographical location information.
3. The short video live broadcasting marketing recommendation method fusing multi-platform behavior features according to claim 1, characterized in that: in the second step, the collected data preprocessing module (2) specifically works as follows:
1) The pre-processing of the store commodity data comprises the processing of commodity price, commodity release time, commodity type and commodity sales volume, and the user store commodity data is preprocessed into
Figure FDA0003867870990000028
The pre-processed commodity sequence of the user shop is
Figure FDA0003867870990000029
2) The preprocessing of the collected data also comprises preprocessing the demographic information of the broadcasting main user to obtain the processed demographic characteristics
Figure FDA00038678709900000210
3) The MySQL relational database is used for processing the latest K of the processed user shop commodities in the persistence processing of the acquired data r A sequence
Figure FDA00038678709900000211
And demographic information
Figure FDA00038678709900000212
And performing persistent storage.
4. The short-video live marketing recommendation method integrating multi-platform behavior features according to claim 1, characterized by comprising the following steps of: in the third step, the recommendation system historical behavior data preprocessing module (3) specifically works as follows:
1) Defining a single historical behavioral data of a live marketing recommendation system as
Figure FDA0003867870990000031
Wherein x is b A single historical behavior data representing the broadcasting of the primary user,
Figure FDA0003867870990000032
denotes the kth b The historical data of the behavior of the user is stored,
Figure FDA0003867870990000033
representing a task budget, representing
Figure FDA0003867870990000034
The time at which the task was issued,
Figure FDA0003867870990000035
indicating that the browsing time of the main user is played,
Figure FDA0003867870990000036
the status of the task is represented and,
Figure FDA0003867870990000037
indicating the number of the task-associated loaded commodities;
2) Selecting the nearest K of the user r Historical behavioral data, deficiency K r The part of the broadcast master user is filled by using a default value to obtain the latest K of the broadcast master user r A historical behavior data sequence K b Is defined as
Figure FDA0003867870990000038
3) Preprocessing broadcast master user recent K r A historical behavior data sequence X b Obtaining a single historical behavior signature after preprocessing as
Figure FDA0003867870990000039
The sequence of historical behaviors is
Figure FDA00038678709900000310
K for processed historical behavioral data sequence using MySQL relational database r A sequence
Figure FDA00038678709900000311
Performing persistent storage of sequences
Figure FDA00038678709900000312
And performing persistent storage.
5. The short-video live marketing recommendation method integrating multi-platform behavior features according to claim 1, characterized by comprising the following steps of: in the fourth step, the specific work of establishing the live broadcast task recall model module (4) is as follows:
1) Define live tasks as
Figure FDA00038678709900000313
t budget For task budgeting, t pos For task release time, t state For the task state, t telate Associating the number of loaded items for the task, t model A task mode is adopted;
2) Input of live task recall model as pre-processed user store merchandise sequence
Figure FDA0003867870990000041
Demographic information
Figure FDA0003867870990000042
And historical behavior sequence of live marketing recommendation system
Figure FDA0003867870990000043
Outputting a sequence of central tasks represented by a marketing task type of interest to a user
Figure FDA0003867870990000044
3) The direct broadcast task recall model calculates the most similar N in the central tasks represented by the marketing task types by using a k nearest neighbor method f Individual tasks forming a set of recalled tasks
Figure FDA0003867870990000045
N h The number of tasks recalled.
6. The converged multi-platform behavioral characteristics according to claim 1The short video live broadcast marketing recommendation method is characterized by comprising the following steps: in the fifth step, the specific work of the live broadcast task database storage module (5) is to collect the task set T obtained in the fourth step h Persisted to a live task database.
7. The short video live broadcasting marketing recommendation method fusing multi-platform behavior features according to claim 1, characterized in that: in the sixth step, the specific work of establishing the live task sequencing model module (6) is as follows:
1) The input of the live broadcast task sequencing model is a user shop commodity sequence after preprocessing
Figure FDA0003867870990000046
And recall tasks
Figure FDA0003867870990000047
2) The live broadcast task sequencing model obtains a predicted click rate, and the tasks are sequenced from large to small according to the click rate to obtain a final recommended task sequence
Figure FDA0003867870990000048
8. The short-video live marketing recommendation method integrating multi-platform behavior features according to claim 1, characterized by comprising the following steps of: in the seventh step, after the broadcasting master user logs in the live broadcasting marketing recommendation system, N with the highest click rate is taken out from the live broadcasting task recommendation database top And recommending the tasks to the broadcasting main user as final marketing recommendation tasks to be recommended to the users on the page.
9. The utility model provides a short video live broadcast marketing recommendation system who fuses multi-platform behavioral characteristics, includes data acquisition module (1), data collection preprocessing module (2), recommendation system historical behavior data preprocessing module (3), establishes live broadcast task recall model module (4), live broadcast task database storage module (5), establishes live broadcast task sequencing model module (6) and live broadcast task recommendation module (7), its characterized in that: the system is characterized in that the data acquisition module (1) is in control connection with a data acquisition preprocessing module (2), the data acquisition preprocessing module (2) and the recommendation system historical behavior data preprocessing module (3) are in control connection with a live broadcast task recall model establishing module (4), the live broadcast task recall model establishing module (4) is in control connection with a live broadcast task database storage module (5), the live broadcast task database storage module (5) is in control connection with a live broadcast task sequencing model establishing module (6), and the live broadcast task sequencing model establishing module (6) is in control connection with a live broadcast task recommendation module (7).
CN202211186906.8A 2022-09-27 2022-09-27 Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics Pending CN115544358A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211186906.8A CN115544358A (en) 2022-09-27 2022-09-27 Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211186906.8A CN115544358A (en) 2022-09-27 2022-09-27 Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics

Publications (1)

Publication Number Publication Date
CN115544358A true CN115544358A (en) 2022-12-30

Family

ID=84729971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211186906.8A Pending CN115544358A (en) 2022-09-27 2022-09-27 Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics

Country Status (1)

Country Link
CN (1) CN115544358A (en)

Similar Documents

Publication Publication Date Title
CN102629360B (en) A kind of effective dynamic commodity recommend method and commercial product recommending system
CN107332910B (en) Information pushing method and device
CN102254028A (en) Personalized commodity recommending method and system which integrate attributes and structural similarity
CN112435067A (en) Intelligent advertisement putting method and system for cross-e-commerce platform and social platform
CN108921602B (en) User purchasing behavior prediction method based on integrated neural network
CN112347362A (en) Personalized recommendation method based on graph self-encoder
CN112307329A (en) Resource recommendation method and device, equipment and storage medium
CN112862530A (en) Marketing system based on big data
CN112511846A (en) E-commerce live broadcast processing method based on big data and network security live broadcast platform
CN104615721A (en) Method and system for recommending communities based on returned goods related information
CN113469730A (en) Customer repurchase prediction method and device based on RF-LightGBM fusion model under non-contract scene
CN114971747A (en) Data analysis method and system based on big data commodity accurate marketing
CN115760202A (en) Product operation management system and method based on artificial intelligence
CN115496566A (en) Regional specialty recommendation method and system based on big data
CN112396496B (en) Intelligent commodity recommendation management system of electronic commerce platform based on cloud computing
CN113570421A (en) E-commerce marketing system based on big data analysis
CN111461846B (en) Shopping record analysis feedback system and method thereof
CN110634051B (en) Fresh agricultural product recommendation method based on multi-granularity fuzzy data
CN115544358A (en) Short video live broadcast marketing recommendation system and method fusing multi-platform behavior characteristics
Arboleda et al. Temporal visual profiling of market basket analysis
CN116091171A (en) Member statistics and management system
CN113283960B (en) Vertical e-commerce platform commodity intelligent recommendation method based on big data analysis and cloud computing and cloud service platform
CN115563176A (en) Electronic commerce data processing system and method
CN113065928A (en) E-commerce transaction method based on big data
CN111427919A (en) Brand feature extraction method and device based on e-commerce recommendation model

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination