Laurent Brisson

How to Semantically Enhance a Data Mining Process

This paper presents the KEOPS data mining methodology centered on domain knowledge integration. KEOPS is a CRISP-DM compliant methodology which integrates a knowledge base and an ontology. In this paper, we focus first on the pre-processing steps of business understanding and data understanding in order to build an ontology driven information system (ODIS). Then we show how the knowledge base is used for the post-processing step of model interpretation. We detail the role of the ontology and we define a part-way interestingness measure that integrates both objective and subjective criteria in order to eval model relevance according to expert knowledge. We present experiments conducted on real data and their results.

Opinion mining on experience feedback: A case study on smartphones reviews

Through the development of electronic commerce, social media and collaborative media, the social commerce appeared. Social commerce, a subset of electronic commerce, is based on social interactions in order to buy and sell goods and services. Nowadays, before buying, people give more importance to the experience feedback they found on internet. However, it is difficult to get an overview of this experience feedback since it is scattered in many online resources, and buyers never have time to read many pages of comments. In this paper, we present an approach which grabs and analyzes experience feedback in order to publish a summary of opinions about a product. We develop this approach with a case study on smart phones and publish a dataset of thousands of comments on a wide range of smart phones. To summarize experience feedback, we use a linguistic appraisal model, based on appreciation, affect and judgement, and we set up an approach using methods and tools from the fields of natural language processing, opinion mining and sentiment analysis.

Challenges to building a platform for a breast cancer risk score

Cancer has recently become the leading cause of death worldwide according to the World Health Organization. As a consequence, health authorities acknowledge the need to implement prevention and screening programs to decrease its incidence. The efficiency of these programs can be increased by targeting higher risk subsets of the population. Efficient tools capable of monitoring the population risk are therefore needed. Constraints to building cancer risk scores and impacts on the tools platform are presented. Major constraints beyond performance of a risk score concern the role of domain experts and their acceptability by end users. Readability is therefore an important criterion. It is shown that a simple k-nearest-neighbor algorithm can achieve good performance with the help of the domain expert. To illustrate this, a risk score made of only four attributes is presented for the French population.

Information Dissemination in Scale-Free Networks: Profusion Versus Scarcity

The study of information dissemination in social networks is of particular importance in many areas as marketing, politics and security for example. Various strategies are being developed to disseminate information, those aimed at disseminating information widely and those aimed at disseminating information in a more confidential manner to make it scarce. In this paper, we adapt a model dedicated to spreading rumours by word of mouth in a physical space to the context of social networks. We compare two modes of dissemination based on profusion or scarcity and study the impact of the choice of the initial node. The results obtained show to what extent each mode exploits the social network topology and especially the influence of hubs.

A Nearest Neighbor Approach to Build a Readable Risk Score for Breast Cancer

According to the World Health Organization, starting from 2010, cancer has become the leading cause of death worldwide. Prevention of major cancer localizations through a quantified assessment of risk factors is a major concern in order to decrease their impact in our society. Our objective is to test the performances of a modeling method that answers to needs and constraints of end users. In this article, we follow a data mining process to build a reliable assessment tool for primary breast cancer risk. A k-nearest-neighbor algorithm is used to compute a risk score for different profiles from a public database. We empirically show that it is possible to achieve the same performances as logistic regressions with less attributes and a more easily readable model. The process includes the intervention of a domain expert, during an offline step of the process, who helps to select one of the numerous model variations by combining at best, physician expectations and performances. A risk score made of four parameters: age, breast density, number of affected first degree relatives and breast biopsy, is chosen. Detection performance measured with the area under the ROC curve is 0.637. A graphical user interface is presented to show how users will interact with this risk score.

Interesting patterns extraction using prior knowledge

One important challenge in data mining is to extract interesting knowledge and useful information for expert users. Since data mining algorithms extracts a huge quantity of patterns it is therefore necessary to filter out those patterns using various measures. This paper presents IMAK, a part-way interestingness measure between objective and subjective measure, which evaluates patterns considering expert knowledge. Our main contribution is to improve interesting patterns extraction using relationships defined into an ontology.