Ehrenfried H.V., Todt E. (2022) Should I Buy or Should I Pass: E-Commerce Datasets in Portuguese. In: Pinheiro V. et al. (eds) Computational Processing of the Portuguese Language. PROPOR 2022. Lecture Notes in Computer Science, vol 13208. Springer, Cham. https://doi.org/10.1007/978-3-030-98305-5_40
- Bibitex
@InProceedings{Ehrenfried_and_Todt_PROPOR2022,
author="Ehrenfried, Henrique Varella
and Todt, Eduardo",
editor="Pinheiro, Vl{\'a}dia
and Gamallo, Pablo
and Amaro, Raquel
and Scarton, Carolina
and Batista, Fernando
and Silva, Diego
and Magro, Catarina
and Pinto, Hugo",
title="Should I Buy or Should I Pass: E-Commerce Datasets in Portuguese",
booktitle="Computational Processing of the Portuguese Language",
year="2022",
publisher="Springer International Publishing",
address="Cham",
pages="420--425",
abstract="Text classification is an essential task in Natural Language Processing. The researchers and developers need data in the desired language to build new models and algorithms to develop this task. In this paper, we discuss and make available three datasets of text classification in Portuguese based on data of the B2W group, which is an important contribution to the research in the field.",
isbn="978-3-030-98305-5"
}
The original dataset was published in the STIL 2019: http://comissoes.sbc.org.br/ce-pln/stil2019/accepted.html
B2W-Reviews01: An open product reviews corpus - Livy Real (B2W Digital); Marcio Oshiro (B2W Digital); Alexandre Mafra (B2W Digital);
Real, L., Oshiro, M., Mafra, A.: B2w-reviews01: An open product reviews corpus. Proceedings of STIL - Symposium in Information and Human Language Technology (2019)
Using the provided dataset three datasets were created. They are B2W-Polarity-Balanced, B2W-Rating-Balanced and B2W-Recommend-Balanced.
The file Dataset.zip contains the dataset without separtion between Train and Test dataset.
The file Experiment_dataset.zip contains the dataset with separation in Train and Test dataset. The Test dataset is composed of 500 reivews of each class of the dataset, while the Train dataset is the remaining reviews
Each file is in the format <class>_<id>.txt. Where <class> is the class of the document and <id> the document identifier. Details about the meaning of each class is described bellow.
In this Section the creation process of each of the datasets is described.
In this dataset, the categories 1 and 2 were grouped into the class Negative (0), the categories 4 and 5 were grouped int othe class Positive (1) and the category 3 was renamed to class Neutral (2).
| Rating | Count |
|---|---|
| 1 | 27369 |
| 2 | 8389 |
| 3 | 16315 |
| 4 | 32345 |
| 5 | 47955 |
The number of reviews are unbalanced across classes, as the following table shows:
| Class | Count |
|---|---|
| Negative | 35758 |
| Neutral | 16315 |
| Positive | 80300 |
| ---------- | ------- |
| TOTAL | 132373 |
In this dataset, classes Negative and Positive are reduced to be contain the same ammount of reviews as the Neutral class. The selection of the reviews are made randomly. Resulting in the dataset described by the following table:
| Class | Count |
|---|---|
| Negative | 16315 |
| Neutral | 16315 |
| Positive | 16315 |
| ---------- | ------- |
| TOTAL | 48945 |
The test dataset is composes of 500 documents from each class.
In this dataset the rating of the reviews are the classes. The Rating 1 is the lowest rate while 5 is the best rate. In this dataset all classes contain the same number of reviews. The removal of reviews in the ratings 1, 3, 4 and 5 were made randomly.
| Rating | Count |
|---|---|
| 1 | 8389 |
| 2 | 8389 |
| 3 | 8389 |
| 4 | 8389 |
| 5 | 8389 |
| -------- | ------- |
| TOTAL | 41945 |
The test dataset is composes of 500 documents from each class.
In this dataset there is 2 classes, the No (0) and Yes (1). The null values are removed from the dataset since it can't be set neither "Yes" nor "No". The original numbers are showed in the table below
| Recommend to a friend | Count |
|---|---|
| No | 35987 |
| Yes | 96368 |
| null | 18 |
Additionaly, the classes were balanced to contain the same number of reviews. Therefore, The "Yes" class lost some reviews randomly, leading to the following final dataset:
| Recommend to a friend | Count |
|---|---|
| No | 35987 |
| Yes | 35987 |
| ----------------------- | ------- |
| TOTAL | 71974 |
The test dataset is composes of 500 documents from each class.
The following table specifies all datasets objectives and sizes
| Dataset | Objective | Balanced | Dataset size | Number of classes | Train dataset size | Test dataset size |
|---|---|---|---|---|---|---|
| B2W-Polarity-Balanced | Classify polarity of the review | YES | 48945 | 3 | 47445 | 1500 |
| B2W-Rating-Balanced | Classify rating of the review | YES | 41945 | 5 | 39445 | 2500 |
| B2W-Recommend-Balanced | Predict if a product is recommended or not | YES | 71974 | 2 | 70974 | 1000 |
The following table presents the statistics of the dataset concernig the number of words per review.
| METRIC | NUMERIC OBSERVATION |
|---|---|
| Mean | 24.296533280956087 |
| Geometric Mean | 18.47804203265341 |
| Harmonic Mean | 14.861573589133467 |
| Median | 16 |
| Mode | 10 |
| Standard deviation | 24.5049628883782 |
| Max Value | 796 |
| Min Value | 1 |
| Number of review with more than 75 words | 4892 |
| Number of review with more than 150 words | 688 |
In the Polarity dataset there is the following reviews
Polarity 0:
Vocês são muito bons para vender, mas entregar é uma outra conversa. A previsão de entrega era para o dia 10/01 e até hoje nada. Estou muito decepcionado com a falta de seriedade de vocês. É muito provável que essa seja a última vez que compro na Americanas.
Polarty 1:
Produto com excelente custo X benefício. Chegou antes do prazo. Parabéns para a Americanas e a LG
Polarity 2:
O sistema de menu da AOC (i-MENU) não está funcionando para mim, eu o instalei, e abri, mas quando vou mexer nas opções apenas 3 é possível clicar, as outras 5 não funciona (Passa o cursor por cima da opção, e clica, mas não funciona), gostaria de saber se é normal nesse produto.