Page 184 - Journal of Library Science in China, Vol.47, 2021

ZHANG Wei, WANG Hao, DENG Sanhong & ZHANG Baolong / Sentiment term extraction 183
and application of Chinese ancient poetry text for digital humanities

1 Related works

1.1 Knowledge organization and humanistic connotation of Chinese ancient poetry

Ancient poetry is an important carrier of our poetic and literary heritage. It contains a wealth of
humanistic knowledge. The current knowledge organization of ancient poetry texts mostly focuses
[7]
on automated classification, whose dimensions include poetic style , poetic subject matter , and
[6]
poetic rhythm . However, most of these studies are in the perspective of the external structure of
[8]
the poetry text, while the excavation and emotional analysis of the inner meaning of ancient poems
are more humanistic. In this regard, WU Bin et al. determined the poems into positive, medium,
and negative emotional polarity based on a migration learning model ; to improve the accuracy of
[9]
emotional classification, TANG et al. proposed a sentiment analysis scheme for poetry of the Tang
Dynasty based on CNNs and GRUs. They adopted a multi-channel processing model to extract
semantic features, and verified the effectiveness of the research scheme by comparing it with
[10]
mainstream deep learning models . However, all the above studies are chapter-level sentiment
polarity analysis, which is difficult to gain insight into the fine-grained humanistic knowledge in
the poems. In contrast, sentiment terminology can more finely analyze the humanistic content of
ancient poetry, but the problems such as limited terminology scale, coarse sentiment granularity
and insufficient learning of text features caused by the condensed style of ancient poetry need to be
[11]
solved first. Appreciation text of poetry is a modern literary text that analyzes the connotations of
ancient poetry and can significantly expand the number of domain terms and refine the emotional
knowledge granularity. Based on the above considerations, we introduce a modern appreciation
approach by integrating the appreciation texts of ancient poetry into the corresponding poems and
devote to the humanistic connotation analysis of word granularity in ancient poetry texts.

1.2 Construction methods of domain sentiment lexicons

Sentiment lexicon is a word granular knowledge organization and sentiment analysis tool . At
[12]
present, the blankness of sentiment lexicon in the field of Chinese ancient poetry makes automatic
construction work imperative. The core construction methods include: 1)Semantic knowledge
base, that is, building an sentiment lexicon on the basis of general sentimental dictionaries (such
as HowNet) by mining word relationships (cognitive and transitive, subordinate) [13] , although
this method is fast and practical, it has the problem of poor domain applicability; 2)Domain
corpus, which aims to automatically extract sentiment word from domain texts, mainly includes
connection relationship method , Statistics (inter-point mutual information, chi-square statistics,
[14]
etc.) [15] and representation learning method (such as Word2Vec) [16] . Among them, the language
rules resorted to the connection relationship method are likely to limit the coverage of sentiment
[17]
word ; statistics are difficult to avoid the inherent defects in the recognition of low-frequency