Cheng-Yu Lu

Emotion Sensing for Internet Chatting: A Web Mining Approach for Affective Categorization of Events

This study proposes an emotion detection engine for real time Internet chatting applications. We adopt a Web scale text mining approach that automates the categorization of affection state of daily events. We first accumulated a huge collection of real-life entities from Web that would participate in events with a user in the chatting room. Based on the common actions between each entity and the type of the user in a chatting room session, such as boy, girl, old man and so on, each collected entity was automatically classified into different affective categories such as pleasant, provoking, grievous, and scary. During a chatting session, each sentence is first parsed using semantic roles labeling techniques to retrieve the verb and object of the event embedded in the sentence. Based on a set of manually authored emotion generation rule, the system then assigns the emotion based on the verb and the affective categories of the object. Primitive evaluations show that the precision rate of the emotion detection engine is rather satisfactory for applications that distinguish emotions of Happiness, Sadness, Anger, and Fear.

Using Web-Mining for Academic Measurement and Scholar Recommendation in Expert Finding System

Scholars usually spend great deal of time on searching and reading papers of key researchers. However, to objectively determine key researcher of a topic relies on several measurements, such as publication, citation, recent academic activities. In this paper, a prototype of scholars searching and recommendation system based on a web mining approach in expert finding system is proposed. The system gives and recommends the ranking of scholars and turns out top-k scholars. A new ranking measure is designed, namely p-index, to reveal the scholar ranking of a certain field. We use a real-world dataset to test the robustness, the experiment results show our approach outperforms other existing approaches and users are highly interested in using the system again.

Mining Fuzzy Domain Ontology Based on Concept Vector from Wikipedia Category Network

Ontology is essential in the formalization of domain knowledge for effective human-computer interactions (i.e., expert-finding). Many researchers have proposed approaches to measure the similarity between concepts by accessing fuzzy domain ontology. However, engineering of the construction of domain ontologies turns out to be labor intensive and tedious. In this paper, we propose an approach to mine domain concepts from Wikipedia Category Network, and to generate the fuzzy relation based on a concept vector extraction method to measure the relatedness between a single term and a concept. Our methodology can conceptualize domain knowledge by mining Wikipedia Category Network. An empirical experiment is conducted to evaluate the robustness by using TREC dataset. Experiment results show the constructed fuzzy domain ontology derived by proposed approach can discover robust fuzzy domain ontology with satisfactory accuracy in information retrieval tasks.

Novelty Paper Recommendation Using Citation Authority Diffusion

Survey of academic literature or papers should be considered with both relevance and importance of references. Authors cite related references by considering integrity and novelty. However, the state-of-art publicly academic search engines and services can only recommend related papers of a certain topic. It shows to manually evaluate the novelty of the recommended papers is necessary. In this paper, we propose a citation-network-based methodology, namely Citation Authority Diffusion (CAD), to rapidly mine the limited key papers of a topic, and measure the novelty on literature survey. A defined Authority Matrix (

Automatic English-Chinese name translation by using Web-Mining and phonetic similarity

AM

Disambiguating authors in citations on the web and authorship correlations

) is used to standardize duplication rate of authors and to describe the authority relation between the citing and the cited papers. Based on

Constructing an Efficient State Space Query System for the Voyage Data Recorder.

AM

Optimal investments strategy for whole life insurance policies

, our

DLConnector: Connecting a publication list to scholarly digital library

CAD

Event-Level textual emotion sensing based on common action distributions between event participants

methodology leverages the Belief Propagation to diffuse the authority among the citation network. Therefore, CAD transforms the converged citation network to a novelty paper list to researchers. The experimental results show CAD can mine more novelty papers by using real-world cases.