CN111598648A

CN111598648A - Full-link online marketing method based on fast-moving industrial commodities

Info

Publication number: CN111598648A
Application number: CN202010300946.5A
Authority: CN
Inventors: 朱传炳
Original assignee: Shanghai Source Hui Information Polytron Technologies Inc
Current assignee: Shanghai Source Hui Information Polytron Technologies Inc
Priority date: 2020-04-16
Filing date: 2020-04-16
Publication date: 2020-08-28

Abstract

The invention discloses a full-link online marketing method based on fast-moving industrial commodities, which comprises the following steps of A, obtaining user information; B. transaction information management is carried out; C. analyzing the interaction behavior data; D. the invention establishes data cleaning ETL and completes data lake full link centralized data service, by using a DTC mode, a brand merchant is directly faced to consumers, the back support is user private domain data and private domain flow, the users are grouped and portrait by some calculation, a fact label and a calculation prediction label are calculated by combining behavior data and service data, the users are more accurately touched, and brand sales conversion is improved.

Description

Full-link online marketing method based on fast-moving industrial commodities

Technical Field

The invention relates to the technical field of network marketing, in particular to a full-link online marketing method based on fast-moving industrial commodities.

Background

The prior art fast-selling goods have the following defects:

1. lack of systematic operation framework and outdated operation team efficiency capability.

Through big and middle platform, the operation module is gathered, and the mode doing work of hot plug can be made up, uses the upper strata to do the polymerization, through unified authentication, all functions of authority management middle platform promote the operation ability.

Decoupling is carried out on the minimum atomic layer in a micro-service mode, application aggregation is carried out on an application layer, data aggregation is carried out on a data layer, the service effectiveness is monitored by a unified service gateway, the service is managed in a unified mode, and the platform supports strong expansibility based on a unified and open API standard.

3. The brand private domain flow closed loop can not be established, the brand private domain data can not be effectively used for on-line touch, and the information has excessive isolated island

The method comprises the steps of establishing a unified bottom data center, centralizing and uniformly collecting data (collecting data through a self-developed tracking system), marking user information by 360 degrees through a self-developed labeling system, and cleaning the data through a data cleaning tool.

4. A full life cycle based private domain membership hierarchy is not established.

And recording all behavior data and transaction data of the user based on the user information base with the unique ID, enriching a user tag system, uniformly storing the user tag system in a brand data lake, and establishing brand private domain data and a member system.

5. It is difficult for a brand to connect to its own private users precisely and individually.

The front-end DTC mall end connection is carried out through an open personalized recommendation engine, real-time strategy rules are provided, for example, a user who browses a certain commodity for 10 times mainly recommends a coupon of the corresponding commodity, and for example, a user who does not visit for more than 7 days actively sends a care notice and the coupon.

And the front end calls the user fact label and the grade in real time through a standard interface to carry out rule matching.

Disclosure of Invention

The invention aims to provide a full-link online marketing method based on fast-moving industrial commodities, so as to solve the problems in the background technology.

In order to achieve the purpose, the invention provides the following technical scheme:

a full-link online marketing method based on fast-moving industry commodities comprises the following steps:

A. acquiring user information;

B. transaction information management is carried out;

C. analyzing the interaction behavior data;

D. and establishing a data cleaning ETL to complete the centralized data service of the data lake full link.

As a further technical scheme of the invention: the user information of the step A comprises user basic information, social attributes and interest attributes.

As a further technical scheme of the invention: the transaction information of the step B comprises user order information, historical purchase information, product preference, browsing preference and conversion channel information.

As a further technical scheme of the invention: the interactive behavior data of the step C comprises browsing behavior, activity preference, behavior attribute and service tendency.

As a further technical scheme of the invention: the ETL in the step D is used for describing the process of extracting, converting and loading the data from the source end to the destination end.

As a further technical scheme of the invention: step A is to acquire two types of data related to a user through user access and a trade order, wherein the two types of data comprise semi-structured data and structured data.

As a further technical scheme of the invention: the semi-structured data comprises user access identity ID, equipment information, browser version information, IP address, OPENID, LBS coordinate information, access time, access page, dwell time and click event; the structured data includes: shopping cart data, receiving information, logistics information, payment information, and order data.

Compared with the prior art, the invention has the beneficial effects that: according to the invention, by using a DTC mode, the back support of a brand merchant is user private domain data and private domain flow, the user is grouped and portrait through some calculations, a fact label and a calculation prediction label are calculated through the combination of behavior data and service data, the user is touched more accurately, and brand sales conversion is promoted.

Drawings

FIG. 1 is a diagram of the data processing architecture of the present invention;

fig. 2 is a schematic diagram of the strategy established by the present invention.

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.

Example 1: referring to fig. 1-2, a full link online marketing method based on fast-moving industry commodities comprises the following steps:

1. acquiring user information: including user basic information, social attributes, interest attributes, and other attributes

2. And (3) transaction information management: user order information, historical purchase information, product preference, browsing preference, conversion channel information

3. Analyzing the interaction behavior data: browsing behavior, activity preferences, behavior attributes, serving trends

4. Establishing data cleaning ETL (abbreviation of English Extract-Transform-Load) for describing the process of extracting (Extract), converting (Transform) and loading (Load) data from a source end to a destination end, and centralizing a data lake whole link

The system management is carried out according to the following steps:

1. push for daily operation

Operation in holidays: drawing corresponding accurate user groups according to browsing habits, shopping characteristic values and the like

Periodic multi-touch content reach (by tag, aggregate grouping, rule, dynamic pull meeting user group for accurate reach

2. Personalized push engine (real-time rule calculation)

Consumer lifecycle personalized push

Real-time response multi-contact push

3. Signature system query

Screening detailed information of users meeting conditions through tags

Querying the user tag through the account: such as purchasing preferences, liveness, etc.

Example 2: on the basis of embodiment 1, based on the management of the full-link and full-life cycle of the user:

1. and (4) carrying out refreshment:

the service is pulled through channels such as coupon dispatching, advertisement, member recruitment, offline activities, online activities and the like;

and (4) pulling the new member to enter a private flow/private member pool to carry out effective member management.

2. Activating:

through the characteristic service, the one-to-one service and the sample sending service, the user pays attention to care to activate the vitality of the members.

3. Interaction

User interaction and user loyalty establishment are carried out through content, grouping, online community, sharing, social fission, live broadcasting and offline activities.

4. Conversion:

user conversion is carried out through modes such as promotion, marketing, accurate touch, limited commodity, brand activity, distribution and the like.

5. And (4) retention:

user life cycle management is performed through loyalty programs, point systems, member points and member rights and interests.

The specific implementation mode is as follows:

the main technology of the project comprises the following steps: the method comprises the steps of large Data acquisition (ES Tracking), large Data preprocessing (ETL), large Data storage and management (Data WareHouse), large Data analysis and mining, large Data display and application (large Data retrieval, large Data visualization, large Data application, large Data safety and the like), mass Data queue cache of distributed storage and sharing of CDN mirror image technology, high concurrency solution and file column storage.

The platform supports various scheduling strategies (time/interface notification/manual/message/short message/WeChat/APP) based on various technical components (such as MapReduce, HBASE, SPARK and HIVE) of Hadoop/SPARK, a relational database storage process, ETL-based calculation, data cleaning, operation through python and the like, and is flexible in transverse capacity expansion based on a micro-service architecture.

The method comprises the steps of forming a user portrait by collecting unique user information through behaviors, orders and transactions, forming a label system, and analyzing, classifying, calculating and regressing through various data models. And analyzing the user access link to form data assets, and outputting accurate calculation and analysis.

Data embedding point based on online user industry

(1) Collecting user behavior information and online and offline service data to form data tracking capable of realizing full link

The data are cleaned, filtered, stored and calculated, the shape of the user is recorded in a full link as a track, the access path and depth of the user are analyzed, and structural shape data and non-structural shape data are integrated to form a data lake with mass data:

data collection based on service level:

(1) and gathering main information, channel information and commodity relevant information of shopping identification SKUs in the items to form a commercial data closed loop of the whole link of the goods yard.

(2) Establishing a massive transaction data summarization mainly based on actual participation in commodity sales, wherein the massive transaction data summarization mainly comprises commodity, marketing, payment, offline, point, user loyalty program, brand and channel related real data sets

The data processing flow and method are as follows:

1. collecting data:

the system collects two types of data related to the user through user access and trade orders:

semi-structured data (behavioral data), mainly comprising:

ID of user access, device information (such as mobile phone model), browser version information, IP address, OPENID, LBS coordinate information, access time, access page, dwell time, click event

The data source is as follows: h5 page, media access, social, brand official web.

Structured data (transactional data) consisting essentially of:

shopping cart data, receipt information, logistics information, payment information, order data (order number, order time, orderer ID, orderer, order time, details of goods corresponding to the order, amount, quantity, terminal page, source, introducer, sales promotion information, discount).

2. Processing data:

the method mainly completes the operations of analyzing, extracting, cleaning and the like of the received data.

1. Extracting: since the acquired data may have various structures and types, the data extraction process can help us convert the complex data into a single or convenient configuration for processing, so as to achieve the purpose of rapid analysis and processing.

2. Cleaning: for large data, it is not all valuable, some data are not the content we are interested in, and other data are the interference terms of complete errors, so the data are filtered and "denoised" to extract the valid data.

3. The private domain data processing method comprises the following steps:

according to the mining method, the method can be divided into a machine learning method, a statistical method, a neural network method and a database method.

The machine learning method adopts inductive learning method (decision tree, rule induction, etc.), case-based learning, genetic algorithm, etc.

The statistical method adopts regression analysis (multiple regression, autoregression, etc.), discriminant analysis (Bayes discriminant, Fisher discriminant, nonparametric discriminant, etc.), cluster analysis (systematic cluster, dynamic cluster, etc.), exploratory analysis (principal component analysis, correlation analysis, etc.), etc.

The neural network method adopts a forward neural network (BP algorithm and the like), a self-organizing neural network (self-organizing feature mapping, competitive learning and the like) and the like.

The database method is mainly a multidimensional data analysis or OLAP (On-Line Analytical Processing) method, and the data storage adopts a HADOOP/HDFS method and an attribute-oriented induction method.

From the perspective of the excavation task and the excavation method, the following steps are mainly adopted:

1. visual analysis: data visualization is the most basic function, whether for the average user or the expert in data analysis. The data imaging can make the data speak by itself, and the user can intuitively feel the result.

Through a big data visualization analysis tool, online dragging, online configuration of flow nodes, dynamic analysis of sales result prediction based on an RMF (recent consumption, frequent consumption) and consumption amount (money) analysis model, and the definition of user classification by recommending commodities with high correlation with customer purchase demand or providing additional repeated purchase reward by adopting a strategy of 'Cross-Sell' (Cross-Sell) or 'Up-Sell'), wherein R, F, M is defined into 5 grades in each direction, 5 is 5 x 5 is equal to 125 types of user classification, and the steps are as follows:

raw data in three dimensions are grabbed R, F, M;

defining R, F, M an evaluation model and a median;

carrying out data processing to obtain R, F, M values;

layering the users according to the evaluation model and the median;

specifying operation strategies for different levels of users;

a data table is finally generated from the sampled data, examples of which are as follows:

the operation strategy is formulated by combining the occupation ratio of various users in the product and the actual business logic of the product. Taking the user hierarchy of a certain shopping guide platform as an example, the strategy shown in fig. 2 is formulated:

helping the company make sales decisions based on the results of fig. 2.

The data visualization technology has 3 distinct characteristics: first, the interactivity with the user is strong. Users are no longer recipients in information dissemination, and can also conveniently manage and develop data in an interactive manner. And secondly, the data display is multidimensional. Under visual analysis, the data sorts, orders, combines, and displays the values for each dimension so that multiple attributes or variables of the data representing an object or event can be seen. And thirdly, the visual visibility is characterized. The data can be displayed in images, curves, two-dimensional graphics, three-dimensional volumes and animations, and their patterns and interrelationships can be visually analyzed.

2. And (3) a data mining algorithm: visualization is the translation of machine language to human, while data mining is the native language of a machine. The analysis of segmentation, clustering and isolated points also has various algorithms of various types of five-flower eight doors, so that the data are refined and the value is mined. These algorithms must be able to cope with large amounts of data while also having a high processing speed.

The data mining algorithm is mainly applied to data mining (KDD) Knowledge discovery in database

Interesting data patterns are found from a wide variety of application data.

The data source includes: databases, data warehouses, the Web, other information repositories.

Types of data that can be mined: database data, data warehouse data, transaction data.

1. Database data:

i.e., data in a database system (also called a database management system: consisting of a set of internally related data, i.e., a database, a set of management, software programs that access the data). The most common is a relational database.

A relational database is a collection of tables, each table consisting of a number of tuples, each tuple representing an object, having a unique identifier (key), and having a number of attributes.

2. A data warehouse:

is a repository of information collected from multiple data sources and stored in a schema at a single site.

The data warehouse is configured to store data by: data cleaning, data transformation, data integration, data loading and periodic data refreshing.

3. Transaction data:

each record of the transaction database represents a transaction, such as a purchase by a customer. A transaction contains a unique identifier ID, and a set of entries that make up the transaction. (e.g., shopping basket analysis (association rules)).

Mining summaries of other types of data;

in addition to the above data, there are other data in various forms and structures. The following were used:

1. time-related or sequence data such as transactions, history, time-series;

2. data streams such as hardware data, sensor data;

3. spatial data such as maps;

4. social data such as comment interaction and comment evaluation;

5. graph and mesh data such as social and information networks;

3. predictive analysis: predictive analysis may allow the analyst to make some prospective decisions based on the results of imaging analysis and data mining.

The result output is mainly performed by the following algorithm:

logistic Regression (Logistic Regression):

logistic regression is a powerful statistical method that can represent a binomial result with one or more explanatory variables. It measures the relationship between a class dependent variable and one or more independent variables by estimating the probability using a logistic function, the latter obeying a cumulative logistic distribution, the main role: credit scoring

Calculating the success rate of the marketing campaign;

predicting revenue for a product;

whether a particular day will consume behavior;

naive bayes classification (Naive Bayesian classification);

naive bayes classifiers are a class of simple probabilistic classifiers that are based on bayesian theorems and strong (naive) independent assumptions between features. The figure is a Bayesian equation where P (A | B) is the posterior probability, P (B | A) is the likelihood, P (A) is the class prior probability, and P (B) is the predicted prior probability.

For example:

judging the junk mails;

classifying categories of crowd labels, such as local tyrant, family boy, shopping madness, and the like;

determining whether the emotion expressed by the text is positive or negative;

face recognition and picture recognition.

4. A semantic engine: the semantic engine needs to be designed with enough artificial intelligence to actively extract information from the data. The language processing technology comprises machine translation, emotion analysis, public opinion analysis, intelligent input, question and answer system and the like.

The deep interpretation semantic recognition engine mainly plays a role:

and (3) emotion analysis: in order to find out the speaker attitude on certain topics, namely information prediction. This attitude may be his (her) personal judgment or assessment, and may be his (her) emotional state at the time.

The technical means for realizing the method are as follows: based on emotion analysis of supervised learning, the supervised learning algorithm used is TFIDF to calculate TfidTransformer and text vectorization CountVectorizer. The emotional tendency of the sentence is judged by scoring the input sentence through a series of dictionaries such as degree adverbs of the emotional words.

Intention recognition: the system has the advantages that the system can accurately position and inquire price, know company conditions, know product conditions, ask for intentions such as contact information and suspicion, and effectively screen out the intended customers in a large number of calls.

The whole process is as follows:

1. obtaining training corpus 2, corpus preprocessing 3, generating word vector 4, training by using LSTM (time-cycle neural network)

2. And respectively adopting LDA document theme models (text classification is carried out by calculating sentence similarity) and improving the identification accuracy by using a feature vector model.

Data quality and data management: the data is processed through standardized procedures and machines to ensure that a predetermined quality of analysis results is obtained. And meanwhile, data is safely stored, and key information is stored by adding salt.

The method mainly comprises the following steps:

1. unifying external data service interfaces to realize all requirements, namely one interface;

2. unifying the caliber of the data index and eliminating the ambiguity of the data;

3. the 360-degree data full link tracking is realized through a data map and a data blood margin;

4. data security encryption storage and hash storage;

5. and multi-source data butt joint, namely butt joint of real-time asymmetric encryption interfaces, real-time data import, database synchronization and the like.

4. Application of data:

and (4) getting through the member interaction link based on a personalized tag system (one person with multiple tags, dynamic tags and an automatic learning function) and rules set by a personalized recommendation engine.

The algorithm is applied: logistic regression algorithm/gradient iterative decision tree algorithm.

Meanwhile, private domain data of a brand are finally generated according to data cleaning and calculation, real-time interaction is calculated in real time through modes such as a small program/H5/APP/API and the like based on a full life cycle data model of the same user identity, and a data label and behavior are triggered according to a rule model, so that a data closed loop is established, and consumers can acquire information interested by themselves accurately according to personalized behavior habits and transaction habits.

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.

Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims

1. A full-link online marketing method based on fast-moving industrial commodities is characterized by comprising the following steps:

A. acquiring user information;

B. transaction information management is carried out;

C. analyzing the interaction behavior data;

2. The full-link online marketing method based on fast-moving industry commodities as claimed in claim 1, wherein the user information of the step a comprises user basic information, social attributes and interest attributes.

3. The method of claim 1, wherein the transaction information of step B comprises user order information, historical purchase information, product preferences, browsing preferences, and conversion channel information.

4. The method of claim 1, wherein the interactive behavior data of step C comprises browsing behavior, activity preference, behavior attribute and service tendency.

5. The fast-moving industry commodity-based full-link online marketing method according to claim 1, wherein the ETL in the step D is used to describe a process of extracting, converting and loading data from a source end to a destination end.

6. The fast food industry goods-based full-link online marketing method according to claim 1, wherein the step a is to acquire two types of data related to the user through user access and trade order, including semi-structured data and structured data.

7. The full-link online marketing method based on fast-moving industry commodities, according to claim 6, characterized in that the semi-structured data comprises ID of user visit, equipment information, browser version information, IP address, OPENID, LBS coordinate information, visit time, visit page, stay time, click event; the structured data includes: shopping cart data, receiving information, logistics information, payment information, and order data.