Overview of Recommender System
Writing On Dissertation: The term web personalization can be explained as a task of constructing web related information systems that are suited for user’s outlook and significance. An efficient customized website should identify its users and based on their expectations and the information should be gathered in attempting to streamline the website content. One of the techniques used in web personalization is a web recommendation system. This system can also be referred as a Recommender System (RS). It is usually a subtype of filtering system that stores collective information.
This system will seek out for predicting the ‘preference’ or ‘rating’, a person, which would be provided to a specific item. Precisely, these systems have been widely employed for users to locate information on the web. These web recommender systems are majorly deployed in commercial applications like rating a place, rating a movie, etc.
There are three types of RS such as Content based, collaborative and hybrid filtering. For example, CF-based systems rely on ratings, whereas content based methods focused on text-based explanations as well as on target user ratings. Therefore, each system has its own strength and weakness. If any of the cases needs a wide variety of inputs, the program will utilize various types of RS in combination.
HYBRID FILTERING RS
The hybrid recommender system (HRS) combines CBRS and CF recommender methods which aids certain limitations of individual.
HRS can be formed in different ways as follows:
- Executing collaborative and methods separately and combining their predictions.
- Integrating some features into a collaborative approach
- Integrating some collaborative features into approach.
- Designing a general unifying model that integrates both content-based and collaborative features.
In this PhD dissertation thesis writing, As mentioned below, all of the above approaches have been used by the recommended systems researchers.
Linking Separate Recommenders
In fact, a single way to develop HRS is to enforce separate collaborative and content-based systems. Then we could have two different conditions. First we can incorporate the outputs acquired from individual RS into a final recommendation. Using either a linear combination of ratings or a voting scheme.
Adding Content Based Feature To Collaborative Approach
At the same time, several hybrid recommendation systems, involving the content collaboration method can focus not only on conventional collaboration strategies. But also on the creation of the content profiles for each user. To evaluate the similarity between two users. This allows to overcome some sparsity related issues of collaborative approaches. Since not many pairs of users should have the proper number of rate items.
Adding Collaborative Characteristics To Content Based Models
The most common approach in this category is to apply a certain dimensionality reduction technique to a group of content profiles. For instance, utilize latent semantic indexing (LSI) to produce a shared view of a set of user profiles. Where user-profiles can define terms vectors, resulting in an enhanced performance compared to a pure content-based approach.
Emerging a Single Unifying Recommendation Model
This approach has been followed in many researchers in recent years. This approach proposes a single unified model by mixing content filtering with collaborative filtering. The recommendation referred to above may be used individually as final recommendations. Or they may combine into a mathematical model to produce a final recommendation for the user. The recommendation will need to be addressed to users after they have produce. The presentation strategy, including what to present. How to present and when to present. It will influence users’ perception satisfaction from the system.
In this PhD dissertation writing, the hybridization thus emerges recently. Through hybridization, various types of RS will mix to attain the best performance. Such hybrid models may closely link to the RS-based ensemble. Many types of algorithms will integrate in this study to create a more robust model. This hybridization principle not only strengthens the use but also improves the performance of specific RS class.