As the saying goes, a clever woman can't cook without rice. Since it is APPPUSH information based on user interest model, there must be user interest model first. So how to build an interest model library for app users?
1, collecting user behavior information
Behavior can reflect interest. The quickest way for a man to catch up with a woman he likes is to pay more attention to a woman's usual behavior, so as to understand her hobbies and finally get what she likes. In the same way, if an app operator wants to make a good user interest model, the first thing he should do is to collect the desired user behavior information through various means.
What is user behavior information?
Any operation of a user on a webpage, website or app product is collectively referred to as user behavior. Taking the shopping app as an example, users' behavior information includes opening the app, logging in an account, searching or browsing goods, joining a shopping cart, paying orders, sharing goods and collecting goods. For another example, the program content watched by the user on a video app, the number of times the user watches a video, and the viewing time are all information about the user's behavior of changing the app product.
How to collect user behavior information
Whether it is a website or an app, there are two common ways to collect user behavior:
Display collection: The new user registration pages of many products explain this way of collecting user behavior information well. Zhihu registration page and Everyone is a product manager registration page. A simple registration page collects several important user information: mobile phone number and name.
In addition to the user's registration behavior, user display feedback (such as uninstalling a software and popping up a web page for you to fill in the feedback information) and form filling can directly obtain user behavior information, which belongs to the category of display collection.
The advantages of this method are: simple and rude, the effect is intuitive, and the collected user behavior information can reflect user preferences to a great extent.
Implicit collection: record all the user's operation behaviors on websites, webpages or apps in the background through technical means, and extract the user's interest information from these operation behaviors.
This is that I added a treasure to the shopping cart in a shop in Taobao one night, and the next morning I received a message reminding me that trading time was running out. The reason why the store's second child sent me this message is because he can see my "joining the shopping cart" in the background. This is a typical case of implicitly collecting user behavior information.
Advantages of this method: users do not directly participate, and the user experience is better.
2. Extract user interest information.
Users visit a certain page and do some operations, which largely represents their hobbies, that is, user behavior information is the real demand of users for products. In addition, we collect so much user behavior data in order to better understand users' interests.
For example, Weibo, WeChat and Momo are all social software, but each user uses them for different purposes, some for obtaining hot news events, some for developing potential customers or opportunities, and some just for strengthening contact with others. This result is produced because each user's interest points are different, so each user's behavior will be different.
Therefore, operators should be able to extract user interest information from user behavior information after they have the data of book use behavior. Suggestions on extracting users' interest information are put forward from two aspects: users' access to content and users' browsing behavior.
With behavioral data, how to extract user interest information and describe user interest in an appropriate way?
3. Display user interest information
After successfully extracting user interests from user behavior information, users' interests need to be described in a more appropriate way. This has two purposes: first, to make the user's interest information more intuitive; The second is to make the later user interest modeling more scientific.
There are three ways to display user interest information:
Theme display method:
The user's interest points are expressed based on the concept set of topic categories. Before downloading Douban app and opening it to the homepage, new users will be asked to choose their own topics of interest.
Advantages: user interest information is intuitive and easy to operate.
Disadvantages: The accuracy of displayed information is open to question.
Keyword display method:
As the name implies, it is through the user's initiative to select some keywords that can represent the user's interests, or adopt relevant algorithms to show the user's interest information. For example, the "personality tag" H5 made by Plato's application can help users to generate a personality tag map, which is the keyword display method used.
Advantages: simple and fast operation.
Insufficient: The user's active participation may be required, and the experience is not good enough.
Vector space representation;
By calculating the weight of keywords, the user's interests are displayed as a feature vector sequence composed of keywords and weights. In other words, vector space representation is an upgraded version of keyword representation, which adds the weight information of keywords and represents the text as a word vector with weight information.
The following figure is the calculation formula of vector space representation.
W(t, d): the weight of the word t in the text d;
TF(t, d): word frequency in text d;
N: the total number of training texts;
Nk: The number of T texts in the training text set.
Advantages: User information can be obtained directly from the server without manual participation.
Disadvantages: the number of text feature words that users are interested in is huge, which increases the workload of operation.
4. Establish a user interest model.
According to the previous user interest information extraction, display, analysis and mining, combined with the actual needs of users, the data dimension type is specified with the user data entity as the center, and the interest model is established for users in a structured form by using the obtained user interest preferences. This link is also often called the establishment of user portraits, that is, marking users and marking user interest information.
Second, how to make users respond to app activities according to the user interest model?
After the user interest model is established, the user interest behavior, user interest tags, user interest portraits and other data about user interest are complete, and the subsequent personalized accurate activity push is relatively simple.
An app can take two forms to push personalized activities according to users' interests. One is to embed the SDK in the app to trigger precise activities according to the actual scenes and needs of users; The second is to actively push the corresponding app activities according to the user interest model.
1.SDK Buried Point Directionally Triggers Personalized Activities
Determine the activity scene and user interest information, determine the activity location, and then embed the SDK in the corresponding position in the app. An activity can only be triggered when the user completes a predetermined operation or meets corresponding conditions (such as activity triggering behavior).
For example, what is the main purpose of an app activity, is it refreshing and retaining the app or promoting the users of the app? Activities in different scenes are placed in different locations of the app. If it is to activate new users, it is necessary to put the active pages that users may be interested in after the registration of app users is completed. Once the user is registered successfully, a new user lottery will pop up immediately.
Another example is a shopping app. If you want to improve the user's repurchase rate, you can put in a coupon activity when the user adds the goods to the shopping cart according to the user's preference for browsing the goods, or put in a discount activity on some surrounding goods according to the user's interest after the user places an order. The picture below shows the trigger coupon activity shared after the takeaway app is hungry.
2. Actively push activities
Active push information intelligently judges what kind of activities users are interested in through the user interest tag in the user interest model database, and then directly recommends the corresponding activities to users. At this point, the mobile news app product that focuses on "interest reading" has done a good job.
For example, daily express, today's headlines and so on. All of them recommend content to users in the way of "intelligent calculation of user interests-+editing operation". The advantage of this is that it conforms to the reading habits and preferences of app users and achieves the effect of personalized reading recommendation.
abstract
Finally, whether it is personalized push of information or personalized push of activities, personalized push is to better realize the process of matching information with users' personalized needs. In this process, it is very important to establish the user interest model, because the user's behavior or interest will change with time or other factors. Therefore, it is recommended that app operators regularly update or modify the user interest model.