2021 SIGIR Group based Personalized Search by Integrating Search Behaviour and Friend Network

[ZahraTaherikhonakdar]

Main Problem:

This paper formed the user's friend networks to personalize the web search. They consider two types of friends:
1- They formed the behavioral-based friends model based on friends who have similar interests like movies or sports etc. (search-log)
2- They formed the relational-based friends model based on friends who share the same experience. (friends network in the social network)
Moreover, they consider mutual friends in both models as more effective friends.
They also create the user profile with the user's search log (considering both long term and short term behavior)

Input-Output:

phase:
Input: friends model+search log+query
Output: personalized ranking documents

Previous Works and their Gaps:
They argue that previous works either consider the search log or friends network to personalized the web search. The shortcoming of the previous works is that they are good for users who have a search history; however, they are insufficient for users who have not used search engines enough or for the new input queries. So, to improve personalized web search performance, they decided to consider users' search logs and users' friends' models.

Result:
The proposed model, FNPS, outperforms the previous personal web search model.

They also, observed that considering more friends would lead to poor performance. Based on the results friends modeling improve the performance for unrepeated queries. Moreover, when the user has enough search history the user profile is sufficient in understanding the user's search intent and friends modeling is useful when the user's search history is poor.

Data Set:
They collected the dataset which contains search logs and friends networks from big social network platform. (they didn't name the platform).

Gap of this work:
This paper is so close to our research interest in terms of the way we look at web search personalization. we could improve this work by considering the user's comments, likes, followings in social networks.

Code:

[hosseinfani]

thanks @ZahraTaherikhonakdar
Try to investigate more on this paper. Very close to our research!