Uyghur–Kazakh–Kirghiz Text Keyword Extraction Based on Morpheme Segmentation

In this study, based on a morpheme segmentation framework, we researched text keyword extraction method for Uyghur, Kazakh and Kirghiz languages, which have similar grammatical lexical structures. these affixes stem are joined together to form word. A is word particle with notional meaning, while the perform functions. Because of derivative properties, vocabularies used languages huge. Therefore, pre-processing necessary step in NLP tasks Kirghiz. Morpheme enabled us remove suffixes as auxiliary unit retaining meaningful it reduced dimension feature space present task texts. We transformed into problem labeling sequences, Bi-LSTM network bidirectionally obtain position information character sequences. applied CRF effectively learn preceding following label sequences build highly accurate Bi-LSTM_CRF model, prepared morpheme-based experimental sets by using model. Subsequently, vectors’ similarity modify TextRank algorithm, subsequent training embedding vector Doc2vec then performed experiment. experiment, highest F1 scores 43.8%, 44% 43.9% were obtained three datasets. The results show that approach provides much better than word-based approach, shows weighting an efficient task, thus proving efficiency sequence morphologically languages.