Construction of knowledge map of mall commodities

Knowledge map construction

The following highlights Alibaba digital commerce Upgrading of knowledge map and related work .

1. Digital business knowledge map upgrade

Under such a large mechanism and model design , The map of digital business knowledge is roughly shown in the figure above . Manage and organize huge business elements through knowledge map . It will be divided into four layers ：

①  first floor Ontology layer , That is, the product knowledge map schema, The main problems to be solved in the data construction of this layer are ：

• how Intelligently and dynamically update schema, Make this tree schema Trees can quickly and efficiently capture 、 Insight into new market trends 、 New changes .

• Each business type of Alibaba has its own products 、 The store CPV, How to build standard schema, So that the commodity data of different markets can be connected , How to do cross market category unification 、 Attribute normalization .

• Previous schema Most designs are Category attribute systems , It can comprehensively and accurately describe the cognition and understanding of the objective part of commodities , But with personalization 、 There is an increasing demand for refinement , How to capture the cognition and understanding of the subjective part of depicting commodities , It is also a difficult and urgent problem to be solved in the future .

②  Below the ontology layer is The conceptual level , Have a certain A class of goods with the same attributes It can be abstractly summarized as a concept , such as fit 「 Appointment 」 The goods , Yes 「 moisturizing 」 Effective goods ,「 Yang Mi speaks 」 Products, etc , It has certain generalization , Interpretability , Abstraction , And there are various relationships between concepts , for example 「 The crowd =0-3 Month Baoma 」 need 「 Size =NB, category = baby diapers 」,「 composition = hyaluronic acid , category = Face cream 」 have 「 efficacy = moisturizing 」 wait , These together constitute the conceptual map . The scientific research community also has the concept of a common sense map , The main function of this layer is to digitize the knowledge or experience that exists in the minds of retail shopping guides or industry operators , Use concepts 、 Concept - The relationship between concepts to describe this knowledge / Experience , The main problems to be solved in this layer are ：

• How to ensure the richness of the concept map , Quickly and efficiently capture the latest elements in the market 、 Concept .

• How to mine efficiently In the concept map 「 Knowledge triples 」, That is to do large-scale knowledge mining .

③  Physical layer , It refers more to every commodity in the specific ecosystem 、 Every shop 、 Every store , The physical scale of this layer is very huge , The scale is at the level of 10 billion , The problems to be solved at this level mainly include ：

• How to ensure such a large scale Massive commodity data And schema layer 、 The mounting of the concept layer is accurate and rich , That is to build commodities / The store profile.

• How to build highly available cross channel commodity relationship services in the face of massive commodity data , Among them, the relationship between the same paragraph is the most important .

④  The event layer is some events in objective life , Including some digital work on the environment .

2. Ontology layer

  A core challenge to be solved in the ontology layer is Improve the degree of standardization . Every business form of Alibaba before , Including Taobao 、 internationalization 、 Local life, etc , There are Independent category attributes To manage , No interaction between , The data is also impassable , So in order to make the goods circulate , We have made some explorations and trial and error , It probably goes through the following stages ：

①  Are independent of each other ： Different Market system Between two Establish a mapping relationship , The advantages of such a scheme are relatively direct , Less information loss , The disadvantage is the high cost , Inconvenient operation and maintenance .

②  Connections ： From all markets schema In the system , Abstract precipitate a relatively standard knowledge system that can meet the basic needs of most businesses , Standard CPV, Then pass the standard CPV System and every channel CPV Construction mapping relationship , The advantage is that compared with the scheme 1, The cost is relatively reduced ,N A market only needs to be established N A mapping relationship is enough , The disadvantage is that the standard is difficult to determine , Great loss of information .

③  share + Personalized customization ： In the scheme 2 On the basis of , It is hoped that in the future, some new markets need to build their own category attribute system , You can refer to the reference standards first CPV, Help business quickly build to 60 branch , then 60 branch -90 Personalized 、 Refined part , Businesses can be customized and extended according to their own needs and market characteristics , This can further reduce the cost of information exchange in various markets , At the same time, maintain the personalized customization and demand of each market .

3. The conceptual level

Before building the concept map , First of all, think clearly about what you want to build The scope of the concept And constraints , Because there are too many abstract concepts in human society , Everything can be conceptualized ,「 sky 」「 White clouds 」「 Table 」「 happy 」 May be a concept , So what concepts are we going to build , To what extent is the construction completed . The answer to this question , The core is to see Scenarios and problems to be solved What is it? .

For Alibaba digital business knowledge map , What you serve The core scenario is still shopping guide , Hope to help The platform better understands commodities , Do it better People and goods match , Therefore, to build what kind of concept map, we must first study Consumer decision theory , To gain insight into the core decision-making nodes of consumers in the purchase decision-making process , Fortunately, traditional retail , Many economies 、 Marketers have quite mature methodology to explain this problem , What we need to do is to digitize this theory 、 Intelligent .

Through to Howard Shea mode Theoretical analysis of consumer purchase decision , We finally model it into the following steps and links ：

• Need to know

Modeling of shopping demand , Who （ Crowd entity ） Where is it （ Location entity ） What time? （ Time entity ） What to do （ Event entity ）.

• information gathering

After having the demand, there may be a demand for shopping , For example, you need to wear a dress on a date , Bring jewelry , A baby needs a bottle 、 Powdered Milk 、 Pacifiers and so on , At present, consumers often disassemble this part through Baidu 、 You know 、 Babytree 、 Friends and other general search or vertical fields app Get to know , Here, from demand disassembly to category , Even subdivide categories （ Attribute item attribute value + category ）, For example, a dress 、 Cut male lipstick 、 Canon camera 、 Silicone pacifier 、 Avocado Green Dress .

• Program evaluation

Here is mainly to choose one of the most suitable products from many categories , For example, you need to buy a camera , Yes 「 Canon 5D2」 still 「 Sony milkshake 7」.

• Buying decisions

If you decide which product to buy , The main problems to be solved are the products on the platform 、sku Too much , How to quickly let consumers choose the most suitable one among the many choices , Here mainly involves some commodities item Cognition and characterization on granularity , For example, some hot topics 、 Service bid 、 Marketing target 、 Qualification bid 、 Price 、 Merchant / Merchant logo .

• Post purchase behavior

This is mainly about evaluation , Need to build Evaluation label system .