Opinions are used to express views and reviews are used to provide information about how a product is perceived. People contributions lie in posting text messages in the form their opinions and emotions which may be based on different topics such as movie, book, product, and politics and so on. The reviews available online can be available in thousands, so making the right decision to select a product becomes a very tedious task. Several research works has been proposed in the past but they were limited to certain issues discussed in this paper. The reviews are collected which periodically updates itself using crawler discussed in our previous work. Further after applying certain pre-processing tasks in order to filter reviews and remove unwanted tokens, the sentiments are classified according to the novel unsupervised algorithm proposed. Our algorithm does not require annotated training data and is adequate to sufficiently classify the raw text into each domain and it is applicable enough to categorize complex cases of reviews as well. Therefore, we propose a novel unsupervised algorithm for categorizing sentiments into positive, negative and neutral category. The accuracy of the designed algorithm is evaluated using the standard datasets like IRIS, MTCARS, and HAR.

Apriori algorithm is one of the most popular data mining techniques, which is used for mining hidden relationship in large data. With parallelism, a large data set can be mined in less amount of time. Apart from the costly distributed systems, a computer supporting multi core environment can be used for applying parallelism. In this paper an improved Apriori algorithm for multi-core environment is proposed. 
The main contributions of this paper are:
•An efficient Apriori algorithm that applies data parallelism in multi-core environment by reducing the time taken to count the frequency of candidate item sets.
•The performance of proposed algorithm is evaluated for multiple cores on basis of speedup.
•The performance of the proposed algorithm is compared with the other such parallel algorithm and it shows an improvement by more than 15% preliminary experiment.