Movie Recommendation System

The objective of this work is to assess the utility of personalized recommendation system (PRS) in the field of movie recommendation using a new model based on neural network classification and hybrid optimization algorithm. We have used advantages of both the evolutionary optimization algorithms which are Particle swarm optimization (PSO) and Bacteria foraging optimization (BFO). In its implementation a NN classification model is used to obtain a movie recommendation which predict ratings of movie. Parameters or attributes on which movie ratings are dependent are supplied by user’s demographic details and movie content information. The efficiency and accuracy of proposed method is verified by multiple experiments based on the Movie Lens benchmark dataset. Hybrid optimization algorithm selects best attributes from total supplied attributes of recommendation system and gives more accurate rating with less time taken. In present scenario movie database is becoming larger so we need an optimized recommendation system for better performance in terms of time and accuracy.


I. INTRODUCTION
It has been vividly seen that recommendation systems has been a major part of consumerism as majority of people relate recommendation systems as a apart of e-commerce sites and shopping destinations. Therefore, recommendation systems find numerous applications when it comes to purchasing or surfing information about any particular product .To substantiate my view, internet sites, shopping websites, electronics etc. are visited by people on daily basis foe meeting their needs. In addition to this, recommendation systems are conspicuously used in guiding people about variety of products that they enquire or they are interested in buying.
The category of the idea is based upon some general issues that may include previous buying behaviour of the customer and demographics of persons. Paul Re-snick and Hal R. Varian, in the year 1997 introduced the term Recommender System'. There were basically two prominent reasons behind finding this particular name as the first and foremost was recommenders could not work properly as the users were unaware of each other's identity, apart from that the other reason was their recommendation also included the interest of the user. Implicit feedback for recommendations was also being provided in combination with the aggregation techniques [2]. To exemplify, collaborative filtering of internet news was done by a system called Group lens, it worked on Two strategies first one was implicit feedback through reading time and explicit feedback through ratings. Furthermore, taking into consideration the privacy concerns of users a new strategy was also recommended and it was pseudonyms and to combine various recommendations weighted voting was used. Not only the efficiency but the cost was also given attention and it was found that maintenance and progress of these systems was hefty and it was important to identify whether the merits outweigh the demerits.

Classification of Recommendation System
Recommendation systems are broadly classified into three categories.  Content-Based Recommendations Recommendations are made solely based on the attributes of the products which the user preferred previously.  Collaborative Recommendations Recommendations are made based on the similarity between the preferences of users and not the content of the products. The recommended items will depend on what other similar users liked.  Hybrid Approach Recommendations are generated based on the resemblances between users and the analogy between products along with their given feedback.

Objective
Keeping these points in consideration following will be our objectives:  To use Neural Network (NN) classification to recommend the movies based on query input  To optimize the parameters of data attributes to achieve higher accuracy using hybrid algorithm which is combination of two powerful optimization algorithms PSO and BFO  To use the overall accuracy as the objective function  Movie lens recommendation dataset will be used for training and testing using NN classifier.

II. PROPOSED WORK
The data used here is movie lens dataset which has a total of 8 attributes to be optimally selected. Dataset description is detailed in next section of this chapter. Out of these attributes, Hybrid Algorithm will choose all those only which contribute more in the accuracy improvement in recommendation. The number of bacteria in Hybrid Algorithm acts as the number of options available for attributes selection at an iteration and their position is the column choice in the data. The whole data is first divided into testing and training datasets. 80% of data is used for training and rest is used for testing. For the first iteration, each position of agent is chosen randomly which is for number of attributes set for an iteration. The selected attributes for them are used to make trained model using NN classifier and tested for the query data. This accuracy is noted down in a matrix for each Hybrid algorithm bacteria. A complete step by step Hybrid algorithm is explained below. 1) Step1: Load the movie lens dataset in numeric format and divide that into random 80:20 ratio for training and testing of recommendation engine.

III. RESULT
We further optimized the attributes for the accuracy improvement using hybrid optimization and compared the results with GA and hybrid algorithm. Any optimization algorithm works well if it converges early and settle to a maximum value (in our case) with no further changes. We plotted iteration curve for hybrid algorithm to check whether our optimization is selecting the optimal attributes or not. Error given by hybrid algorithm (combination of PSO and BFO) is also very less as comparison to GA algorithm. Other values such as prevalence , specificity, positive likelihood, positive predictive value and error are also indicate that results produced by hybrid algorithm is convincing than GA. We used neural network classifier to recommend the movies similar to user selected movie. The multiclass here because the rating of movie is the criteria on which NN modeling will recommend the movie and this rating is in between 1-5.  Here it is clearly evident that accuracy of getting correct rating of movie using proposed hybrid algorithm is better 0.7969 than accuracy obtained from GA which is 0.614.  Here it is clearly evident that sensitivity using proposed hybrid algorithm is less which is better (0.085) than sensitivity obtained from GA which is 0.1316.  Here it is clearly evident that specificity using proposed hybrid algorithm is almost same as that of GA.  Here it is clearly evident that positive likelihood using proposed hybrid algorithm is less which is better (0.7267) than positive likelihood obtained from GA which is 1.    .1(f) shows the bar chart of prevalence obtained from both the algorithms. Here it is clearly evident that prevalence using proposed hybrid algorithm is same as that of GA.
For better comparison results class-wise, confusion matrix is generally used to plot. So, we used confusion matrix plots to analyze the accuracy performance by each class. Confusion matrix will tell us how many samples were picked for each class and exact number of samples detected in each class after model testing. Figure 4.2 shows the confusion matrix plot for these two optimizations. The yellow highlighted box in confusion map represents the accuracy % for that class.  Yellow diagonal elements shows truly predicted output labels by hybrid algorithm. It is observed that for class 1,2,3,4,5 truly predicted ratio are 72.6 %,65.52 %,100 %,85.9 %,59.43 % respectively. It means average predicted values are 77% which is very good.  Yellow diagonal elements shows truly predicted output labels by hybrid algorithm. It is observed that for class 1,2,3,4,5 truly predicted ratio are 100 %, 0 %, 100 %, 61.83 %, 2 % respectively. It means average predicted values are 52% which is less in comparison to our proposed algorithm. Also there is very low value in prediction of two class labels which are class 2 and class 5.

IV. CONCLUSION
In this work a comprehensive study of personalized recommendation system (PRS) in the field of movies recommendation has been carried out. In order to perform the above-mentioned investigation, we have used standard Movie Lens dataset downloaded from web link to create training model using new regression based model. The data is in qualitative form which is converted to quantitative to use in recommendation model. We have proposed a new regression model which is based on neural network (NN) classifier and hybrid optimization algorithm which is combination of BFO and PSO algorithm. We have 1 million user's rating for 1682 movies which is used to make a training model but firstly we have retained important attributes using hybrid algorithm optimization and these reduced features are then utilized to create training model using NN classifier and this model is tested using priory separated testing data and applied to trained model to find out output label accuracy.  [20] Include users behaviour along with movies Addition of new data bias the trained system 21.
Personalized recommendation system (PRS) and Support vector machine (SVM) classification Amarjit kundu,et al. [21] Extracted the un-important features of the data Use PSO based local optimisation for this purpose which has problem of premature termination

22.
Swarm mining Sajal Halder, et al. [22] Help producers to plan a movie Unable to find the group of a user who likes multiple movies depending on the movie genre.

23.
Classical Vector Space Model (CVSM) Roberto Mirizzi, et al. [23] Developed face-book application which link movie recommendation with user's preference Focused on the cold start problem NA