# Visual Data Mining for Collaborative Filtering: A State-of-the-Art Survey

**Authors:** Marenglen Biba (University of New York Tirana, Albania); Narasimha Rao Vajjhala (University of New York Tirana, Albania) — ORCID 0000-0002-8260-2392; Lediona Nishani (University of New York Tirana, Albania)
**Type:** Book chapter
**Source:** In Collaborative Filtering Using Data Mining and Analysis, pp. 217–235, IGI Global
**Published:** 2017
**DOI:** https://doi.org/10.4018/978-1-5225-0489-4.ch012
**Canonical page:** https://www.narasimharao.net/research/visual-data-mining-collaborative-filtering-survey/
**Indexing:** Scopus

## Summary

**Question.** What visual data mining techniques and tools exist, what are the main collaborative filtering approaches, and how has visual data mining been applied to collaborative filtering recommender systems?

**Finding.** The chapter surveys visual data mining taxonomies (for example Keim’s six classes: graph-based, geometric projection, icon-based, pixel-based, hierarchical and hybrid), popular tools (Clementine, TimeSearcher, ThemeRiver, VizTree, XmdvTool), and the three main kinds of collaborative filtering (memory-based, model-based and hybrid). Reviewing visual collaborative filtering systems such as NEAR, dependency networks, minimum spanning dendrograms and PCA/CCA item maps tested on MovieLens and Netflix data, it concludes that combining visual data mining with collaborative filtering has led to substantial improvement and can alleviate the shortcomings of traditional collaborative filtering.

**Why it matters.** Visual data mining augments, rather than replaces, traditional data mining by bringing the data analyst into the exploration process through visual interpretation of large datasets. For recommender systems, the authors conclude that visual techniques help make sense of what is happening between users in collaborative filtering systems and address problems such as sparse data, scalability and new users or items.

## Key findings

1. The survey reports that visual data mining techniques can be grouped into six classes — graph-based, geometric projection, icon-based, pixel-based, hierarchical and hybrid — following Keim’s taxonomy.
2. The chapter identifies Clementine, TimeSearcher, ThemeRiver, VizTree and XmdvTool as popular visual data mining tools, describing Clementine as one of the three most popular visual data mining tools currently in use.
3. The survey classifies collaborative filtering techniques into three major kinds: memory-based, model-based and hybrid collaborative filtering.
4. The chapter notes that memory-based collaborative filtering is easy to implement and performs well on dense datasets but depends on user ratings, deteriorates on sparse data and requires more memory than model-based techniques.
5. The survey identifies the key challenges for collaborative filtering recommender systems as managing highly sparse data, scaling to large numbers of users and items, and producing appropriate recommendations quickly, along with the new-user and new-item (cold-start) problems.
6. Reviewed visual collaborative filtering approaches include the NEAR (navigating exhibitions, annotations and resources) panel, dependency networks, minimum spanning dendrograms for movie recommendation, and PCA- and CCA-based global and personalized item maps tested on MovieLens and Netflix datasets.
7. The authors conclude that combining visual data mining techniques with collaborative filtering has led to substantial improvement of the overall systems and can alleviate the shortcomings of traditional collaborative filtering.

## Study at a glance

| Item | Detail |
|---|---|
| Research question | What are the state-of-the-art visual data mining techniques and collaborative filtering approaches, and how are they combined? |
| Design | State-of-the-art literature survey (book chapter). |
| Data | Published visual data mining taxonomies, tools and collaborative filtering studies, including work evaluated on EachMovie, MovieLens and Netflix rating datasets. |
| Methods | Narrative review of visual data mining classifications, tools, memory-based, model-based and hybrid collaborative filtering, and visual collaborative filtering systems. |
| Main result | Combining visual data mining with collaborative filtering has substantially improved overall systems and can alleviate the shortcomings of traditional collaborative filtering. |
| Implication | Visual data mining should be used alongside traditional data mining so analysts and users can visually explore and interpret patterns in recommender data. |

## Abstract

This book chapter provides a state-of-the-art survey of visual data mining techniques used for collaborative filtering. The chapter begins with a discussion on various visual data mining techniques along with an analysis of the state-of-the-art visual data mining techniques used by researchers as well as in the industry. Collaborative filtering approaches are presented along with an analysis of the state-of-the-art collaborative filtering approaches currently in use in the industry. Visual data mining can provide benefit to existing data mining techniques by providing the users with visual exploration and interpretation of data. The users can use these visual interpretations for further data mining. This chapter dealt with state-of-the-art visual data mining technologies that are currently in use apart. The chapter also includes the key section of the discussion on the latest trends in visual data mining for collaborative filtering.

## Key terms

- **Visual data mining:** The chapter’s definition: visual data mining combines data mining methods and computer-aided, interactive visual techniques to discover novel and interpretable patterns with the help of human perception abilities.
- **Collaborative filtering:** The chapter’s definition: collaborative filtering allows recommendations to users taking into account how other users have rated items, using a database of user ratings over items.
- **Recommender systems:** The chapter’s definition: software tools and techniques for suggesting items to users by considering their preferences in an automated fashion.

## How to cite

Biba, M., Vajjhala, N. R., & Nishani, L. (2017). Visual Data Mining for Collaborative Filtering: A State-of-the-Art Survey. In Collaborative Filtering Using Data Mining and Analysis (pp. 217–235). IGI Global. https://doi.org/10.4018/978-1-5225-0489-4.ch012

```bibtex
@incollection{biba2017visual,
  title = {Visual Data Mining for Collaborative Filtering: A State-of-the-Art Survey},
  author = {Biba, Marenglen and Vajjhala, Narasimha Rao and Nishani, Lediona},
  booktitle = {Collaborative Filtering Using Data Mining and Analysis},
  pages = {217--235},
  year = {2017},
  publisher = {IGI Global},
  doi = {10.4018/978-1-5225-0489-4.ch012},
  url = {https://doi.org/10.4018/978-1-5225-0489-4.ch012}
}
```