Page 159 - JOURNAL OF LIBRARY SCIENCE IN CHINA 2018 Vol. 42

158 Journal of Library Science in China, Vol. 8, 2016

4 Conclusion and discussion

In this paper, we categorize method knowledge elements into five types, i.e. method definition
knowledge element, method feature knowledge element, method feature knowledge element,
method process knowledge element, method functionality knowledge element. We use the semi-
automatic method to construct description rules for these five method knowledge elements.
Examples of these rules are also given. The types and description rules will benefit technology
implementations of method knowledge element extraction in the future.
In this study, we find out that different knowledge elements use different descriptions, the
sentence complexity as well as the types of description vary greatly. The method definition
knowledge element is simple, usually takes one sentence. The sentence rule is simple too. Method
feature knowledge elements describe advantages and disadvantages a lot while neutral descriptions
are rare. Comparative sentences are frequently used when talking about advantages and
disadvantages in one-to-one comparison and one-to-many comparison. The syntactical rules are
comparatively complex. Relations between methods include static relation in space and dynamic
relation in time. There exist only a few static relations but the description rules are complex
while there exist a lot of dynamic relations but the description rules are simple. Meanwhile, it
is difficult to explicitly represent certain relations like substitution in a sentence or a paper; thus
the construction of rules is also difficult. Although we can use conjunction words to aid in the
recognition of method process descriptions, most of the description will require more than one
sentences. Sentence group or paragraphs are more suitable for this type of description and thus the
method proposed in this paper fails to construct rules accordingly. Method functionality description
can be categorized into domain application, problem-solving, etc. Usually, the description requires
only one sentence and thus the construction of rules is not difficult.
The research in this paper is preliminary, and there are still problems and weaknesses. On the
one hand, there is no test dataset for method extraction and thus we cannot evaluate the results.
There is no standard for the number of rules required for each type of method knowledge elements.
On the other hand, because of the complexity of languages and the different styles of paper writing,
generalization ability and the coverage of rules are still in question. In the future, we will increase
the size of the original corpus and construct a test dataset to evaluate the performance.

References

Gao, J. P., Ding, K., Pan, Y. T., & Yuan, J. P. (2015). Review of the study on knowledge elements (知识元研
究述评). Information Studies: Theory & Application (情报理论与实践)，(7),134-138,133.
Gao, Y., Lu, X., Li, N., & Chen, S. F.(2003). An adaptive rule extracting algorithm in probabilistic plan (一种
自适应概率规划规则抽取算法). Journal of Nanjing University (Natural Sciences) (南京大学学报(自然
科学版)), (2),145-152.
Hou, G. K., & Zhang, J. F.(2000). A rule extraction method based on decision tree (基于决策树的神经网络