Co-occurrence Weight Selection for Word Embeddings to Enhance Test Performance

Uludağ University Journal of The Faculty of Engineering

卷:23

Veysel Yücesoy¹  Aykut Koç¹ 

[1] ASELSAN;

关键词: kelime temsilleri;    doğal dil işleme;    i̇statistiksel dilbilimi;    word embeddings;    natural language processing;    statistical linguistics;   

DOI  :  10.17482/uumfd.318615

来源: DOAJ

【 摘 要 】

This study revisitsthe problem of maximizing the performance of mathematical word representationsfor a given task. It is aimed to improve performance in analogy and similaritytasks by suggesting innovative weights instead of the counting weights usedconventionally in counting-based methods of generating word representations(adding the statistics of word co-occurrences to the account). The language ofstudy was selected as Turkish. The root structures of Turkish words were managedduring the compilation of corpus such that each word having a suffix wasconsidered as a new word. The performance of the proposed co-occurrence weightsare analyzed with respect to the varying parameter and the results arepresented within the paper.

【 授权许可】

Unknown