What is a user-portrait?
Along with big data technology development and update, mobile phone user information channel has undergone great changes. Membership management, surveys, shopping cart analysis compared the traditional line, big data technology for the first time enables companies to obtain broader user behavior information via the Internet conveniently for further accurate, rapid analysis of user behavior, consumption habits and other important business information, provides adequate data base. With the gradual deepening of understanding of the people, a concept born quietly: user portrait.
Many people want to know, what is in the end user portrait? About user defined portrait, we must mention is Alan Cooper presented the concept of persona:
Establish target user model on a set of real data. Through user research to understand the user, based on differences in their goals, behavior and opinions, they will be divided into different types, and each type is typically extracted features, giving name, description, photos, a number of demographic factors, scenes, etc. , to form a character archetype.
How to build user-portrait?
The core part of the user is actually the user portrait label included therein, user tags can be understood as a series of symbols in fact, the user characteristic representation, each tag can be understood as a user's understanding of an angle. User portrait is actually a collection of tags, there is some connection between each label, the whole dimension of each label combination to form a complete user portrait together. So the user can actually picture label collections to represent.
Characterized in identifying a tag usually highly refined artificial and labels such as age: 18 to 35 years, regional markers: Shanghai, the label exhibits two important features: first, semantic, one can easily understand that each meaning of the label. It also allows users to model portraits have practical significance. Can better meet business needs; second, short text, each tag usually means only one meaning, the labels themselves do not need to do too much text analysis, preprocessing, which has facilitated the use of standardized information extraction machine.
People develop labeling rules, and can be read quickly by a label which information the machine easy to do tag extraction, aggregation analysis. Therefore, the user portraits, namely: users tag, to show us a simple, concise method for describing user information.
1. Preparation and Analysis of the data source
For now, Internet products, it needs to be done is to have the hands of structured data and cleaning, removal of noise and irregular data ready for subsequent modeling.
What if the product is not particularly small amount of molding or the user how to do it? This is not a particularly good way, a small amount of data, it can only continue to increase the amount of data. There are two ways to get data, the first method is free, there are some published data, companies can collect and downloaded from the Bureau of Statistics Web site or open a website; the second is the payment method, companies can pay to get user data industry, such data obtained is generally more accurate. In addition, companies can also do some actual user research (can be outsourced), to obtain user data research.
Data sources can be divided into static data and dynamic data.
Static data is relatively stable in the user information, as shown, including demographic attributes, business attribute data and so on. Since this type of information into the tag, if there is a real business you do not need too much information modeling predicts that more data is needed cleaning.
Dynamic information refers to information about users' changing behavior. When the feedback consumer behavior on the Internet, and even electricity supplier, will focus a lot of user behavior, as shown below: Where the customer browser home page, browsing casual shoes single product page, search for canvas shoes, shoe-quality micro-blog published about Chan "double XI promote to force" microblogging messages. And so can be seen as Internet user behavior. On the Internet, user behavior can be seen as a dynamic user unique data source information.
2. User Classification
By analyzing the user's static data, users can perform basic categories, such as gender, the user is divided into male and female users, according to the resident city, users can be divided into first-tier cities, second and third tier cities, etc. In addition, also to be divided according to age of the user, user sources, income levels and occupation.
After the static data acquisition, populations need to factor and cluster analysis, based on different classification for different purposes: as for product designs, divided according to use motivation or behavior is the most common way, and for the type of media for marketing, based on consumption patterns to distinguish between people is the most direct classification.
3. User defined and the right to re-label
With static and dynamic after the data, we will be marked with labels for each user, as well as the right to re-label.
Weight how to define it? Sometimes, the weight is defined by the experienced marketing staff to set, they are more understanding of the users or the general situation of the industry, and therefore be defined still have high credibility through experience.
Of course, this weighting may change over time, users in different life cycles, as well as their different behavior will change in the length of heavy impact on this right. In the model, the weights need to constantly optimize at a later stage.
Of the population defined in terms of characteristic values, help to grasp at a glance the characteristics of the group. Refers to the characteristic value is a very distinctive feature of words to describe a group of people, such as "fashionistas", "movement of people" can make people very quickly and clearly know what kind of people are characterized Yes.
4. Data Modeling
Data modeling to solve the thing is, how the user behavior for the user to automatically add labels, set the weight, how to measure the similarity of different users through user behavior, how to make specific recommendations by a similar measure, which may be user data reaches a certain size after things need to be considered.
Now commonly used collaborative filtering recommendation thought, that is, after the user behavior data analysis and quantitative similarity measure, from automated decision out.
Of course, not the only recommendation systems, precision of analysis and decision-making, personal management, and even automated operations are built on this foundation.