I. Project Source Code
Project source code URL: (This project is not open source at present.)
II. Introduction
The same as user profile, the commodities actually have their own set of separate features. By calculating multidimensional features of commodity, it can establish a hierarchy system, basic commodity profile. Such multidimensional features include gender, age, category levels, product word, brand word, hot sale, price, level of purchasing power, etc.
There are also connections among commodities. Under each categories, the relevance of commodities, the topic model clustering and hierarchical abstraction can be done by different dimensions of brand word, product word and extended attribute. It is another hierarchy system, but also belongs to the field of commodity profile. The most typical application of it is that the relevant commodities can be aggregated as different clusters named topics or scenes, that is, personalized recommendation.
III. Basic Commodity Profile
Commodity gender: some commodities are generally sexual distinguishable, for example, some specific handbags are for men and some are for women, which can reflect gender preferences. Of course, there are still some commodities without gender distinction.
Commodity age: some commodities have age range, for example, the shoes include baby shoes, teenager shoes, adult shoes, elderly shoes, etc., which can reflect age preference. But some commodities are without age distinction.
Commodity category: the categories divided according to macro levels.
Product word and Brand word: They are basic attributes of commodity, including specification parameters.
Price and Purchasing power level: A mass consumer product or a high-level consumer product, and the level of consumption.
IV. Extended Commodity Profile
Relevant commodity: the similarity and correlation of commodities can be obtained from user’s daily behaviors. These commodities often combined together must have correlation degree with each other. The higher the correlation degree is, the more likely that the attributes of the related commodities can be used as subsidiary features of the dominant commodity.
Aggregated commodity sets: Clustering by FP-growth or topic models, and candidates are recommended for each other in the same topic set. This is described in detail in “The application of scene recommendation”.
Hierarchical abstraction: Hierarchical subordination among commodities.
V. Reference
[1]…
[2]…
[3]…