For a while I have been thinking about what context in data science exactly means, because it may mean so many things. Quite often we use context as a joker word or a variable (because it is vague) to describe many things. This happens especially then, when we are not sure what those many things are.
So what is query context or user context in Web search? We can say, everything about the query or the user. And then what is everything? For me, the only more concrete bit there is the about part, which implies some attribute, i.e. something that is external. For example, weather forecast is a query, which is in English language. Hence, it has a language attribute, with a defined attribute value English. So, language is about a query, therefore it is an attribute, and therefore it is a part of context data. Context can be a quite subjective concept too, in that, what I understand as context can be different than yours .
Recently, I came across to a blog post, which nicely discusses what context is, and I thought it was quite helpful to have a bit better understanding of it. Here is the original post:
The author discusses it from a product and customer point of view. So, when reading replace product with query and customer with user. Accordingly, context data has three compositional parts: demographic data + historic data + situational data.
Demographic data (this is about ‘who’ the user is):
- user preferences
- his/her query patterns (long queries, short queries, questions, use of numericals…)
- behaviors (how long does s/he dwell, how often quick back, how often and quickly does s/he change to another search engine)
- what other digital channels s/he uses (Twitter, Instagram, Facebook,…)
Historic data (concerns past records of user interaction):
- the set of earlier queries
- whether or not they were satisfied (click vs. quick-back)
- type of queries (e.g. popular, unconventional, multimedia…)
This data also acts as a way to predict the behavior and future actions of users.
Situational data (this data is about the ‘now’):
- the current geographic location of the users
- devices used by them
- other current online activities as can be inferred by the preceding/succeeding links e.g. login to FB (social), Amazon (commercial), Twitter(social), Courserea (education) etc.
This data helps gain an idea of what are the current preferences of the user are, or what are they looking for at a particular time.
Hoping this helps one or the other to have a bit more well-defined understanding of what context may mean, ideally in many other contexts that are quite different than mine.
No comments:
Post a Comment