선택한 단어 수는 입니다.
최소 단어 이상 선택하여야 합니다.
최소 단어 이상 선택하여야 합니다.
선택한 단어 수는 30입니다.
최대 10 단어까지만 선택 가능합니다.
최대 10 단어까지만 선택 가능합니다.
가설 설정
- (2) doc2Vec_Key represents document embedding through doc2Vec with using keywords as dictionary. (3) Proposed SV is the result of generating a document vector by averaging keyword vectors. It should be noticed that in the above three methods(1) ~ (3), every document is represented by a single vector.
제안 방법
- In this paper, we firstly summarize the previous study on multi-vector document embedding(박종인·김남규 2019), and then we visually analyze the effects of the multi-vector document embedding method.
- Therefore, recently, we proposed a new multivector document embedding method(박종인·김남규 2019) to overcome these limitations of the traditional document embedding methods represented by doc2Vec. Firstly, to solve the limitations that a document vector is affected by many meaningless words, the proposed method vectorizes a document using only the word vectors of keywords in the document instead of using the entire document. Secondly, the proposed method separates keyword groups by clustering and then generates multiple subject vectors for each group to vectorize complex documents more accurately.
- Firstly, to solve the limitations that a document vector is affected by many meaningless words, the proposed method vectorizes a document using only the word vectors of keywords in the document instead of using the entire document. Secondly, the proposed method separates keyword groups by clustering and then generates multiple subject vectors for each group to vectorize complex documents more accurately.
- Chapter 4 reveals the results of applying the multi-vector document embedding method to the vectorization of real documents. In particular, the effects of multi-vector embedding are analyzed by visually comparing the results of single vector representations and multi-vector representations. Finally, Chapter 5 summarizes the contributions and limitations of the proposed method.
- , (3) Keyword Vector Extraction, (4) Keyword Clustering, and (5) Multiple Vector Generation in [Figure 2]. The proposed method embeds the document using only the vector of the keyword word, not the whole corpus. The word vectors shown in [Figure 3(b)] includes not only keywords but also vectors of meaningless words.
- In this experiment, a total of 3,147 academic papers were collected on seven categories from the web site. The experiments were performed using Python 3.6, mainly using Komoran for tokenization, Gensim for word2Vec modeling, and Numpy package for vector operations. This section introduces an overview of the experiment, and later sections of this chapter describe the main processes and results of the experiment.
- By eliminating the interference among subjects of complex documents, we ascertained that complex documents can be vectorized more accurately. The proposed method can perform more accurate text structuring, therefore, we expect that multi-vector embedding can improve the quality of various text analysis results such as classification, clustering,and topic modeling.
대상 데이터
- In this experiment, a total of 3,147 academic papers were collected on seven categories from the web site. The experiments were performed using Python 3.
- To more clearly compare the performance of various document embedding methods, the complex documents were artificially created using the category information of each document. The documents used in the experiments are abstracts of 3,147papers in seven categories such as Humanities,Social Sciences, Natural Sciences, Engineering, Agriculture, Medical Science, and Arts. With seven categories, 21 cases of two category combinations(e.
- According to the methods described above, 1,050 complex documents were generated by random sampling and a center-based method. The 350 original documents used to construct the complex document were excluded from the document set. As a result, the word vector was trained on 3,847 documents (2,797 single category documents + 1,050 complex documents) using the word2Vec algorithm.
이론/모형
- This chapter revisit multi-vector document embedding method(박종인·김남규 2019) that we proposed recently with simple examples.
- All identified tokens are structured as a vector with real value through word embedding. In this study, tokens are converted to vectors using the word2Vec algorithm, which is widely used to word embedding methods. An example of converting the tokens to an N-dimensional vector can be seen in [Figure 3].
- For example, in the case of , if the keyword set can be separated into two groups,the proposed method embeds the document as two member vectors. Several techniquescan be used to group keyword vectors, but the proposed method uses the K-means, one of the most representative clustering algorithms. For example, if is represented as a single vector using the average of the keywords, the result will be = (-0.
- The 350 original documents used to construct the complex document were excluded from the document set. As a result, the word vector was trained on 3,847 documents (2,797 single category documents + 1,050 complex documents) using the word2Vec algorithm. The dimension of the vector is 300 dimensions, the learning count is 50, and the window size is set to 5.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.