<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1202">
  <Title>Using Gene Expression Programming to Construct Sentence Ranking Functions for Text Summarization</Title>
  <Section position="6" start_page="2" end_page="2" type="concl">
    <SectionTitle>
5 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper, we have presented a prototype summarization system which employs GEP as its learning mechanism for sentence ranking function. In the preliminary experiments for performance testing, our system outperforms the baseline methods by 58%-160% when generating summaries for 10 documents. However, the value of the average similarity gained by our system is not as high as we would like. The reason most likely lies in the fact that the styles of the objective summaries written by humans vary a lot or even conflict with each other.</Paragraph>
    <Paragraph position="1"> In other words, they do not possess many common features that are a must for high value of similarity between two texts. Using content-words and the cosine function to measure the similarity may not be an ideal evaluation metric, neither is it an ideal fitness measure in the GEP learning mechanism. Our future  research will further study what kinds of similarity measure can be obtained from raw texts without involvement of human subjects. Moreover, we plan to cluster collected documents to make every cluster contains articles summarized in a similar style. We will also explore other sentence features, such as sentence cohesion, semantic meaning, and rhetorical relations, for an ideal uniform sentence ranking function.</Paragraph>
  </Section>
class="xml-element"></Paper>