📜 ⬆️ ⬇️

User interest assessment

At the moment, for more or less large projects there is a rather topical issue of assessing the interests of the audience. That is, it is not a necessity, you can do without it, but on the other hand it can significantly improve the services provided and increase the income from the project.
There is not a lot of information on the Internet on this subject, those who have already made this mechanism quietly rejoice in themselves and are not particularly in a hurry to share their thoughts. And the question is quite complicated. I would like to raise this topic here and try to bring together various ideas. Yes, in many ways the mechanism may depend on the structure of the project, but still the main ideas can be identified.
And so, below I will set out my thoughts on this subject and I will be glad if you leave your comments and new ideas in the comments. If the topic is interesting to readers, in the next post I will try to collect all of the above in the comments.



First, let's imagine the resource itself, where we will evaluate the interests of users:

Users can fill out their interests in the questionnaire. Almost all types of materials can / should be left tags.
User interest assessment should respond to various user actions. In the simplest case, we will represent interests as words or phrases — essentially tezhe tags — and not divide them into groups. That is, for each user we compose a set of interests (tags) and an indicator of “interestingness” of the corresponding tag.
The indicator of “interestingness” can be expressed in the form of a probability or a certain number. I suggest that this indicator be called the “weight” of interest, and stored as a number. If necessary, this weight can be easily converted into probability. The weight of interest increases all the time, i.e. when performing actions by the user, we add to the corresponding weight of interest a certain number.
And so, for each user we have a set of tags - interests - each of which corresponds to a certain number - weight.
Now you need to highlight the actions in which there will be a definition of interests. To start:
  1. Data filled in the account questionnaire.
  2. The user wrote a post on his blog. As an interest, you can use the tags left to this post, you can also index the text of the post itself, but it is more complicated.
  3. Leave a comment on the post or material. We use tags of post or material as interests.
  4. Looking through the material. As an interest tag material.
  5. View photos, videos, listening to audio. Tags to materials. Here you have to be careful, if the audio / video is a song / clip, it’s not a fact that they should be taken into account in the interests (unless you need to define musical preferences), but if this is some kind of teaching material, or advertising, then tags go for the interests.
  6. Site search. If there is a search form on the resource, then everything that the user is looking for will completely come down as an interest.
  7. Joining the community. Use community interests (tags).

These are linear actions that a user can perform on a site. If the topic is interesting, then in the following articles we will talk about more complex user actions.
Now directly operations with scales of interests. It is impossible to evaluate each action equally. Something more weighty, but something less. Therefore, it is necessary to make a list of all possible actions and determine the corresponding plus weights for them.
Here I would like to note that it would be more optimal not just to give each action its plus weight, but to make a certain relationship between them.
Those. choose one action, I called it “equivalent”, and express all the others in relation to it. For example, I propose, as an equivalent action, to use the writing of a post and take into account tags to it. I have designated the equivalent action EQ. Further, we can assume that writing a post is equivalent to 100 comments on a post. Denoting the added weight when adding a comment to a post as CPI (comment post interest), we get CPI = EQ / 100. And so on, in the end we will only have to assign a certain number for EQ, and the rest will be calculated automatically. Thus, you can quickly change the plus weights for all actions, and specify only for specific ones.
I think for a start it is enough, in the following posts I will tell about more complex dependencies of consideration of interests. In the comments I would like to see your additions about the actions performed by the user and about the system of weights.
')

PS This is not the whole idea, this is only its beginning, so that it is not necessary to immediately criticize the plane.

Source: https://habr.com/ru/post/52456/


All Articles