Transfer Pretrained Sentence Encoder to Sentiment Classification

Bai, Man; Han, Xu; Jia, Heng-Yue; Wang, Cong; Sun, Yawei

doi:10.1109/dsc.2018.00068

Cited by 1 publication

(6 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Regarding individual method results, for lexical, similar to ternary classiication, Word -grams overlap ( 1 = 0.74) outperformed VSM-BoW ( 1 = 0.68) and VSM-CNG ( 1 = 0.69). Once again the best result is obtained using a combination of N-gram features [1][2][3][4][5] (5 features) and the Jaccard similarity measure. This shows that the combination of features is helpful in improving performance even in the binary classiication task.…”

Section: Results and Analysismentioning

confidence: 99%

“…łlo-vsm-bowž, łlo-vsm-c4gž, and łlo-vsm-c5gž refers to the Vector Space Model method applied with Bag of Words, Character 4-grams and Character 5-grams, respectively. łsm-lcsž refers to the Longest Common Subsequence while łsm-gst-cmbž refers to the Greedy String Tiling method applied by combining the MML length [1][2][3][4][5] (5 features). łss-sno-j-cmbž and łss-sno-d-cmbž refers to the Stop-word -grams overlap method with N-grams of length [1][2][3][4][5] (5 features) and similarity measures Jaccard and Dice, respectively.…”

Section: Methodsmentioning

confidence: 99%

“…łsm-lcsž refers to the Longest Common Subsequence while łsm-gst-cmbž refers to the Greedy String Tiling method applied by combining the MML length [1][2][3][4][5] (5 features). łss-sno-j-cmbž and łss-sno-d-cmbž refers to the Stop-word -grams overlap method with N-grams of length [1][2][3][4][5] (5 features) and similarity measures Jaccard and Dice, respectively. łwe-w2v-ct-aež refers to the custom trained monolingual Word2Vec model with averaged embeddings method.…”

Section: Methodsmentioning

confidence: 99%

“…Each of these models outputs a ixed-length sentence embedding vector on a given input sentence of any length. Using these models, the degree of similarity between a source and derived text document is computed as follows: All the sentences 4 from each source and derived text document are converted to their respective sentence vectors using one of the sentence embedding models [28]. These sentence embedding vectors are then summed to produce the document-level vector representation.…”

Section: Mono-lingual Word Embeddingsmentioning

confidence: 99%

“…It is a neural network-based model, trained on 570k human-written English sentence pairs from the Stanford Natural Language Inference (SNLI) corpus [8]. The model has recently found success in sentiment analysis [4], text summarisation [16], and question-answering systems [10] tasks.…”

Section: Mono-lingual Word Embeddingsmentioning

confidence: 99%

See 4 more Smart Citations

Cross-lingual Text Reuse Detection at Document Level for English-Urdu Language Pair

Sharjeel

Muneer

Nosheen

et al. 2023

ACM Trans. Asian Low-Resour. Lang. Inf. Process.

View full text Add to dashboard Cite

In recent years, the problem of Cross-Lingual Text Reuse Detection (CLTRD) has gained the interest of the research community due to the availability of large digital repositories and automatic Machine Translation (MT) systems. These systems are readily available and openly accessible, which makes it easier to reuse text across languages but hard to detect. In previous studies, different corpora and methods have been developed for CLTRD at the sentence/passage level for the English-Urdu language pair. However, there is a lack of large standard corpora and methods for CLTRD for the English-Urdu language pair at the document level. To overcome this limitation, the significant contribution of this study is the development of a large benchmark cross-lingual (English-Urdu) text reuse corpus, called the TREU (Text Reuse for English-Urdu) corpus. It contains English to Urdu real cases of text reuse at the document level. The corpus is manually labelled into three categories (Wholly Derived = 672, Partially Derived = 888, and Non Derived = 697) with the source text in English and the derived text in the Urdu language. Another contribution of this study is the evaluation of the TREU corpus using a diversified range of methods to show its usefulness and how it can be utilized in the development of automatic methods for measuring cross-lingual (English-Urdu) text reuse at the document level. The best evaluation results, for both binary ( F 1 = 0.78) and ternary ( F 1 = 0.66) classification tasks, are obtained using a combination of all Translation plus Mono-lingual Analysis (T+MA) based methods. The TREU corpus is publicly available to promote CLTRD research in an under-resourced language, i.e. Urdu.

show abstract