The term "big data" refers to a variety of data types, including text, images, videos, graphics, and more. With a Graphics Processing Unit (GPU)-based framework, the training time for such a model may be reduced from many weeks to only a few hours. Numerous activities and a growth in information processing have resulted in a data overflow. Unstructured data comprises 80% of all organized data. Only completely unstructured, relevant sources should be examined and used. An essential aspect of web data mining is determining which pages are accessed together during a single network visit using association rules. The purpose of Education Data Mining (EDM) is to provide tools for the analysis of particular categories of data generated in a learning setting. In online usage mining, association criteria are used to identify web sites that are linked together via associations rather than via direct connections. Apache Spark is a framework for data processing that uses several computers to distribute processing over very large datasets. Much of the difficult work needed for large data processing and distributed computing is abstracted away by Spark’s API. The web scraper takes the required links from the web, extracts the information from the source links, and stores the information in a csv file. Web scraping is not illegal, but it might not be ethical. Finding trends in teachers’ and students’ system usage is one of the key goals of educational data mining. One of the numerous characteristics of Educational Big Data (EBD), which is generated from the activities of many objects in various learning situations, is complexity. Our research aims to develop and implement a hybrid model based on web mining and recommendation systems for educational big data applications. We’ll start by discussing web data mining approaches (K-means and association rules, for example) and selecting one or more techniques for recommendation systems. To complete the services provided by that educational application, one of the environments or platforms for handling big data is selected.
The increase in data from modern sources, the heterogeneous nature of data, ambiguous and unstructured data, and the so-called Big Data with all of its five v’s characteristics, indicate a growing need to use approaches that provide assistance in modeling and processing these data, provide additional automated data processing, and so on. The majority of all these studies involve one or more big data sets that may be used in various applications. Most of these selected papers use one or more distributed frameworks (such as MapReduce, Spark, or HDFS Hadoop distributed file systems). Furthermore, more than one strategy of Web mining is dealt with (such as Naive Bayes, Logistic Regression, and Random Forest, as examples) more than recommender system types (collaborative filtering, or/and content-based filtering). Various data sets were used (such as Movie Lens data set, LDOSCoMoDa data set, real bank loans dataset, using Facebook and Twitter for collecting data, data was taken across different companies, thousands of movies were used as exemplary data sets, or data sets used a sample data from different schools of Central Board of Secondary Education (CBSE) across India). Testing accuracy on the dataset obtained is required in the accuracy used (precision and recall). The best precision was 0.9886 and the best recall was 0.9835 in these studies. We have displayed some studies that have high precision and low recall and some other studies that have low precision and high recall. This paper introduces a literature survey about recommender systems that deal with big data through web mining techniques.