<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2808">
  <Title>Anomaly Detecting within Dynamic Chinese Chat Text</Title>
  <Section position="10" start_page="53" end_page="54" type="concl">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> The new approaches to detecting anomalous Chinese chat text are proposed in this paper. The approaches calculate confidence and entropy values with the language models constructed on negative training samples in three standard Chi- null nese corpora. To improve detection quality, we incorporate positive training samples in NIL corpus in our approaches. Two conclusions can be made based on this work. Firstly,  F measure of anomaly detection can be improved by around 0.10 when NIL corpus is incorporated into the approaches. Secondly, performance equivalent to the best ones produced by the approaches in existence can be achieved stably by incorporating the standard Chinese corpora and the NIL corpus. We believe some strong evidences for our claims can be obtained by training our approaches with more chat text corpora which contain chat text created in different time periods. We are conducting this experiment seeks to find out whether and how our approaches are independent of time. This work is still progressing. A report on this issue will be available shortly. We also plan to investigate how size of chat text corpus influences performance of our approaches. The goal is to find the optimal size of chat text corpus which can achieve the best performance.</Paragraph>
    <Paragraph position="1"> The readers should also be noted that evaluation in this work is a within-domain test. Due to shortage of chat text resources, no cross-domain test is conducted. In the future cross-domain test, we will investigate how our approaches are independent of domain.</Paragraph>
    <Paragraph position="2"> Eventual goal of chat text processing is to normalize the anomalous chat text, namely, convert it to standard text holding the same meaning. So the work carried out in this paper is the first step leading to this goal. Approaches will be designed to locate the anomalous terms in chat text and map them to standard words.</Paragraph>
  </Section>
class="xml-element"></Paper>