A thesis submitted in fulfilment of the requirements for the degree of Master of Science Computer Science in the School of Computer Science. Faculty of Science
November 21, 2016. / There is stiff competition for customers and market share in the South African telecommunications
industry amongst the four predominant mobile service providers, namely Vodacom,
MTN, Cell C and Telkom Mobile. The First National Bank (FNB) through one of its entities,
FNB Connect, has also joined this intensely competitive environment. These companies face
a constant challenge of having to come up with new and innovative ways of attracting new
customers and retaining their current ones. Cell C has embarked on a good strategy of claiming
solid market share and growing itself against the competition by using the Private Label
Promotions (PLP) group, a leading BEE Level 3 company that provides a variety of business
solutions, to market GetMore, its value-added service package. A recommender system could be
used to suggest and promote the items available in this package to existing and potential clients
(users). There are different approaches to recommendation, the most widely used ones being
the collaborative and content-based recommendation. The collaborative filtering approach uses
the ratings of other users to recommend the items the current (active) user might like. In the
content-based approach, items are recommended in terms of their content similarity to items a
user has previously liked, or elements that have matched a user’s attributes (features). Hybrid
recommendation approaches are used To eliminate the drawbacks individually associated with
the CF and CBF approaches and to leverage their advantages. One of the aims of this research
was to design and implement a prototype hybrid recommender system that would be used to
recommend Cell C’s GetMore package to current and potential subscribers. The system was to
implement matrix factorisation (collaborative) and cosine similarity (content-based) techniques.
Several experiments were conducted to evaluate its performance and quality. The metrics used
included Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Area Under the
ROC Curve (AUC). We expected the proposed hybrid recommender system would leverage the
advantages provided by its different components and demonstrate its effectiveness in providing
Cell C’s customers with accurate and meaningful recommendations of its GetMore package
services.
Keywords:
Content-based Recommendation, Collaborative Recommendation, Hybrid Recommendation,
Cosine Similarity, Matrix Factorisation, Association Rule Mining, J48 Classifier, Decision Table,
Naive Bayes, Simple K-means, Expectation Maximization, Farthest First, Predictive Apriori / LG2017
Identifer | oai:union.ndltd.org:netd.ac.za/oai:union.ndltd.org:wits/oai:wiredspace.wits.ac.za:10539/21650 |
Date | January 2016 |
Creators | Ndlovu, Mpumelelo |
Source Sets | South African National ETD Portal |
Language | English |
Detected Language | English |
Type | Thesis |
Format | Online resource (xv, 221 pages), application/pdf |
Page generated in 0.3736 seconds