Efficient Long Short-Term Memory-Based Sentiment Analysis Of E-Commerce Reviews Part 3

4. Results

We trained our model for about 10 epochs and calculated the training and validation loss as well as training and validation accuracy. 

Loss may seem like an inevitable event in our lives, but its impact on our mental health and memory can be profound.

First, when we experience a loss, it's easy to get caught up in negative emotions. This may include sadness, anger, anxiety, and more. These emotions can affect our mental health, leaving us feeling exhausted, powerless, and disappointed. At the same time, these negative emotions can also affect our memory, making it difficult for us to concentrate and our memory becomes blurred.

Click know supplements to improve memory

However, even when we experience loss, we can still maintain a positive attitude and optimistic perspective, helping us overcome the impact of negative emotions on our mental health and memory. This includes spending time with friends and family, sharing feelings, maintaining a healthy diet and lifestyle, engaging in physical exercise, and attending psychotherapy, among other things.

Additionally, we can take proactive steps to help ourselves maintain good memories during the loss process. This includes giving yourself some time to accept and adapt to new situations, as well as trying new memory techniques and strategies, such as building memory associations, repeating exercises, using memory sticks, etc.

All in all, loss does take a toll on our mental health and memory, but we can take proactive steps to overcome these effects. If we can maintain a positive attitude and give ourselves some time to adjust to new situations, we may be able to more easily regain our mental health and memory and move on to a better life again. It can be seen that we need to improve memory, and Cistanche deserticola can significantly improve memory because Cistanche deserticola is a traditional Chinese medicinal material that has many unique effects, one of which is to improve memory. The efficacy of minced meat comes from the various active ingredients it contains, including acid, polysaccharides, flavonoids, etc. These ingredients can promote brain health in various ways.

We can see from Figure 4 that both the training and validation loss decreased throughout the training of the model. Figure 5 shows that the training and validation accuracy increased subsequently for 10 epochs.

Since, after prediction, the final output we get is a probability, we apply a certain threshold to determine whether the data belongs to the positive or negative class. For this purpose, we have used the ROC curve that plots the true positive and negative rates. 

It helps find the threshold values for a binary classifier. From our ROC curve shown in Figure 6, we have chosen 0.78 as our threshold.

+e saved model is reloaded and predictions have been generated on the test data considering the abovementioned threshold value. Now we have the original sentiment as well as the predicted sentiment. 

Since the dataset is imbalanced, the better parameter to test the model would be the F1 score rather than accuracy. 

In Table 4, we have compiled the accuracy, precision, recall, and F1 scores of other baseline models and compared them with our model. +e baseline models were considered from the literature we reviewed for this experiment.

5. Conclusion+is paper discusses sentiment analysis in the context of e-commerce reviews. There have been various techniques surveyed previously in the field of opinion mining of reviews. 

Our database consists of reviews from the cell phone and accessories section of Amazon. Long Short-Term Memory Networks were used to classify the sentiment using deep learning. Our custom training dataset was used to extract the features embedded in the word2vec embedding technique. Based on the ROC curve, we determined that 0.78 is the final threshold we should use to classify sentiment. 

Four parameters have been used to evaluate our model's performance: accuracy, precision, recall, and F1 score. A precision of 97% is found to be the highest of the four parameters. 

As the dataset is unbalanced, we consider the F1 score as the best measure of the model's performance, which yields an evaluation of 93%. The main attempt of this research was to test the functionality of the model with a large amount of data. +is method provides good results even for such large data of about 938,261 reviews. +e The main advantage of using this method is that LSTM takes into consideration long-term memory and word2vec efficient estimation of word representations which help in efficient sentiment analysis. 

For future work, we would like to consider using bidirectional LSTM for sentiment classification which trains two strands of LSTMs, the actual input sequence and the reverse one. +is might help improve the model performance.

Data Availability

+e data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest

+e authors declare that they have no conflicts of interest.

Acknowledgments

+is research was funded by Princess Nourah bint Abdulrahman University Researchers Supporting Project number. (PNURSP2022R120), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

References

[1] B. Liu, Sentiment Analysis: Mining Opinions, Sentiments, and Emotions, Cambridge Univ. Press, New York, NY, USA, 1st edition, 2015. 

[2] P. Balaji, O. Nagaraju, and D. Haritha, "March). Levels of sentiment analysis and its challenges: a literature review," in Proceedings of the International Conference on Big Data Analytics and Computational Intelligence (ICBDAC), pp. 436–439, IEEE, Chirala, India, March 2017. 

[3] R. Varghese and M. Jayasree, "A survey on sentiment analysis and opinion mining," International Journal of Renewable Energy Technology, vol. 2, no. 11, pp. 312–317, 2013. 

[4] C. Sindhu and D. S. Chandrakala, "A survey on opinion mining and sentiment polarity classification," International Journal of Emerging Technology and Advanced Engineering, vol. 3, pp. 531–539, 2013. 

[5] A. Jurek, Mulvenna, and Y. Bi, "Improved lexicon-based sentiment analysis for social media analytics," Security Informatics, vol. 4, no. 1, p. 9, 2015. 

[6] S. Zhang, D. Zhang, H. Zhong, and G. Wang, "A multiclassification model of sentiment for E-commerce reviews," IEEE Access, vol. 8, pp. 189513–189526, 2020. 

[7] J. Devlin, M. Chang, K. Lee, and K. Toutanova, "BERT: pretraining of deep bidirectional transformers for language understanding," in Proceedings of the Conference North American Chapter Association Computational Linguistics, Human Language Technol, pp. 4171–4186, Minneapolis, Minnesota, June 2019. 

[8] L. Dey, S. Chakraborty, A. Biswas, B. Bose, and S. Tiwari, "Sentiment analysis of review datasets using naive Bayes and k-nn classifier," 2016, https://arxiv.org/abs/1610.09982. 

[9] M. R. Huq, A. Ali, and A. Rahman, "Sentiment analysis on Twitter data using KNN and SVM," International Journal of Advanced Computer Science and Applications, vol. 8, no. 6, pp. 19–25, 2017. 

[10] B. S. Lakshmi, P. S. Raj, and R. R. Vikram, "Sentiment analysis using deep learning technique CNN with KMeans," International Journal of Pure and Applied Mathematics, vol. 114, no. 11, pp. 47–57, 2017. 

[11] Y. Fang, H. Tan, and J. Zhang, "Multi-strategy sentiment analysis of consumer reviews based on semantic fuzziness," IEEE Access, vol. 6, pp. 20625–20631, 2018. 

[12] B. Shin, T. Lee, and J. D. Choi, "Lexicon integrated CNN models with attention for sentiment analysis," 2016, https:// arxiv.org/abs/1610.06272.

For more information:1950477648nn@gmail.com