2016 - Springer - Recommender Systems by Aggarwal, Charu C

[karan96]

Title: Recommender Systems Author: Aggarwal, Charu C Year: 2016 Venue: Springer

Chapter: - 9. TIME- AND LOCATION-SENSITIVE RECOMMENDER SYSTEMS

Section: - 9.4 Location-Aware Recommender Systems

The author explains that any recommender systems that takes into account location can be considered a special case of context-aware recommeder systems. The author explains few of the common ways in which location affects the recommendation process: -

1. A case where location is considered to be associated with the user but not with the item, in our case a team containing that specific user/location. In this case, the users are spatial, whereas the items are not. This is called preference locality.
2. Travel Locality where the location is associated with the item than the users. Users can specify their desire of location which we can use to limit or search space. Hence in this case items are spatial and not the users.
3. A use case where both, the user and the location, are spatial.

Key Point - Usage of multidimensional model by treating location as a context, associating a hierarchical taxonomy of grid regions with the spatial location, and then reducing the problem to a traditional collaborative filtering application within one of the hierarchical regions of the grid.

Preference Locality:

• Similar to reduction-based multidimensional model of recommeder systems. Fixing a location to a certain regioin then extracting a slice of Candidate X Skill X Location. Then a 2-dimensional recommeder system can be used on this.
• LARS approach to partition the data space to combat higher degrees of granularity in location by making a pyramid tree and dividing the space into H levels with each grid cell containing a collaborative filtering model only for that region of data space.

Travel Locality:

• Locations are associated with items and not the users.
• Location associated with the skills and a user might specify their current location for a query to predict responses that are close or within the specified location in the query.
• Achieved in LARS using a travel penalty.

• Combination:

• The recommeder system must also take into account the transient query location of user from a different location trying to find user with skill in different location.
• To use methods mentioned in both preference and travel locality.
• First, the pyramid-tree structure is used, based on the primary user location, in order to predict the ratings. Then, the transient query location is used in conjunction with the aforementioned travel penalty. The top-ranked items are then returned to the user.

[hosseinfani]

@karan96 The book is not recent but it gives you a good direction. I would suggest finding recent papers in each of the categories that this book introduces. Then linked them here and start reading them to come up with a method for our own project.