Efficient Long Short-Term Memory-Based Sentiment Analysis Of E-Commerce Reviews Part 1

In today's modern era, e-commerce is making headway through the process of bringing goods within everyone's grasp. Consumers are not even required to step out of the comfort of their homes to buy things, which makes it very convenient for them. 

With the continuous development of modern technology and the rapid development of e-commerce, our lives have become more convenient and efficient, which has also affected our memory to some extent. However, there is indeed a relationship between e-commerce and memory, and the relationship is positive and upward.

First of all, e-commerce provides us with a more convenient shopping method. We no longer need to go to the mall to shop in person. With just a few clicks of the mouse, the goods we need can be delivered to us by express delivery. This form of shopping eliminates the worries of long waits and crowds, saving us time and energy. With the development of e-commerce, we can use more intelligent technologies to shop, such as voice shopping, intelligent recommendations, etc. These technologies can make our shopping more convenient and efficient.

Secondly, e-commerce can also help us better manage information and data, which is beneficial to our memory and work efficiency. We can use tools such as email, cloud disks, and online notes to record and share important information, freeing our brains. These tools help us better organize and manage information, ensuring our brains only need to focus on the most important things. In this case, our memory is still trained because we use these techniques and tools to help us remember the information rather than trying to remember it.

Finally, e-commerce can also help us learn and develop, which is very beneficial to our memory and intellectual development. We can learn new knowledge and skills through online courses, e-books, online learning platforms, and other tools. These tools allow us to better manage and master knowledge, while also helping us better develop our professional and personal lives. This way of learning can stimulate our thinking and improve our memory and creativity.

Overall, there is indeed a relationship between e-commerce and memory, but it is a positive and upward one. We can use e-commerce to improve our quality of life and work efficiency, and we can also use it to improve our intelligence and memory. Therefore, we should actively use these technologies and tools to add more positive energy to our lives and career development. 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.

Click know 10 ways to improve memory

Moreover, there is a wide variety of brands to choose from. Since more customers depend on online shopping platforms these days, the value of ratings is also growing. To buy these products, people rely solely on the reviews that are being provided about the products. 

To analyze these reviews, sentiment analysis needs to be performed, which can prove useful for both the buyers and the manufacturer. ­is paper describes the process of sentiment analysis and its requirements. 

In this paper, the Amazon Review dataset 2018 has been used for carrying out our research and Long Short-Term Memory (LSTM) has been combined with word2vec representation, resulting in improving the overall performance. 

A gating mechanism was used by LSTM during the training process. ­e proposed LSTM model was evaluated on four performance measures: accuracy, precision, recall, and F1 score, and achieved overall higher results when compared with other baseline models.

1. Introduction

Communication has played a key role in boosting social relationships since historical times. Nowadays, nearly every segment of society uses social media as it has evolved into an efficacious networking tool. ­e main part of social media comprises e-commerce sites. 

Because of the rapid advancement of e-commerce technologies, the majority of people now choose to purchase online. People can use social media to provide feedback on various situations, items, and resources, which can be positive or negative, based on the customer's experience. 

Unfavorable comments play an essential role in the growth of the company because they help to improve the services. Here, sentiment analysis comes into play.

Sentiment analysis aids in giving away a customer's viewpoint on different goods via text information and at the same time assessing these reviews shared. 

Various types of research suggest that sentiment analysis is generally conducted at three levels: sentence, document, and phrase-level [1–3]. ­e substeps involved in the process of sentiment analysis are depicted in Figure 1.

­Is research proposes the use of LSTM networks to classify a large number of Amazon reviews. ­is deep learning technique is fast and gives better results even for a large number of reviews. ­e study uses word2vec embedding for the efficient estimation of word representations in vector space. 

Word2vec provides better results than the standard representation methods such as bag of words or one-part encoding. ­is study mainly focuses on two parts: Efficient mapping of sentiment words into vector space through the word2vec model and the LSTM network to classify reviews.

2. Literature Survey

+is section contains all important background work on the subject of sentiment analysis that is relevant to our research. 

We discovered that most of the earlier works employed machine learning algorithms, deep learning algorithms, and sentiment lexicon. In Table 1, we have summarized the approaches used in the research and the merits and demerits of the approaches.

In the year 2013, Sindhu and Chandrakala [4] observed recent and efficient techniques that are employed to study sentiment analysis, including sentiment polarity classification and various machine learning techniques such as Naive Bayes, Maximum Entropy, and support vector machine. +e survey suggests that sentiment classification can be determined by two attributes, polarity assignment, i.e., determining if the sentiment is positive, negative, or neutral, and intensity assignment, which depicts how strong or mild a particular sentiment is in terms of polarity. 

Jurek et al. [5] presented a model with a lexicon-based sentiment analysis algorithm that included two key components: evidence-based integration function and sentiment normalization that measured emotion rather than a positive/negative label and aided in the differentiation of different emotions. 

A publicly available Twitter Corpus was used as a dataset for this study, the main focus of the study being real-time Twitter content analysis.

Zhang et al. [6] presented a multiclassification approach to perform sentiment analysis on e-commerce reviews. 

Further, Zhang et al. [6] presented a multiclassification model for sentiment analysis of e-commerce reviews. +e Amazon review dataset (2018) was used for the proposed study, which was based on a directed weighted problem. +e proposed study stated that, by extraction of entity words with features, assessment of sentiment patterns, and evaluation of the shortest path between nodes, the problem of sentiment similarity could be transformed into a problem of shortest path computation. When compared to the BERT model [7], this model performed better in terms of the algorithm's CPU time.

Dey et al. [8] examined the machine learning algorithms, K-NN and Naive Bayes, using three evaluation metrics. +e Naive Bayes classifier outperformed the K-NN classifier in their work. 

Researchers in [9] presented a sentiment classification model with two techniques. +e first proposed approach, the sentiment classification algorithm employed the K-NN classifier and in the other one, the support vector machine algorithm was used. +e efficiency of the classification algorithm was validated based on real tweets. +e results obtained showed that the sentiment classification algorithm outperformed SVM on experimental validation. 

In [10], a comparison of supervised and unsupervised learning methods was presented. +eir work provided a comparative analysis of supervised (CNN and KNN) and unsupervised (CNN with K means clustering) learning algorithms.

Fang et al. [11] introduced a multisentiment analysis technique that heavily incorporates fuzzy set theory, machine learning theory, and a polarity lexicon-based method. Consumer reviews were then analyzed using this hybrid model. 

Naive Bayes and SVM algorithms were used for this study. +e enhanced SVM model, i.e., a hybrid method that combines multistrategy sentiment analysis with the SVM, was much more successful and gave an accuracy of 86.35%. 

Additionally, a 3.8% increase in accuracy was observed while implementing the upgraded Naive Bayes. In addition, researchers in [12] presented a way for incorporating lexical embeddings and an attention mechanism into CNN. +e dataset was created using tweets. +e method was evaluated using the F1 score. +e work that was suggested performed better than the present ones.

A Recursive Neural Network (RNN) based recommendation system (RDSA) was introduced by Preethi et al. [13]. 

Deep learning was utilized to optimize suggestions centered on sentiment analysis and was done on three separate reviews in this study. 

Firstly, datasets were investigated and their statistical aspects were observed before implementing the Naive Bayes classifier and the RNN. +e results of the trials showed that using RNN, a deep neural network, boosted the accuracy of sentiment analysis, leading to finer suggestions for users and aiding in the selection of a particular position depending on the requirements of the users.

Furthermore, researchers in [14] proposed using a Gini index-based feature selection and an SVM classifier to categorize data. +e dataset for this study was a large collection of movie reviews. 

Based on the findings of the experiments, the proposed method was determined to be less accurate than other methods. A gated RNN with inter-opinion connections was introduced by Chen et al. [15]. +is approach had an accuracy of about 92.6%. 

For classification, a bidirectional gated recurrent unit (BiGRU) paired with an attention mechanism was proposed in [16]. +is approach was found to be effective for classification tasks and generated better outcomes than previously utilized methods, with a 93.1% accuracy. A replacement sentiment analysis model that incorporates the CNN and the attention-based BiGRU was proposed by researchers in [17]. 

By integrating the benefits of sentiment lexicon with deep learning technology, compensates for flaws in the traditional sentiment analysis model for product reviews. +e sentiment lexicon supports the sentiment attributes found in the reviews and CNN used in conjunction with the gated recurrent unit network extracts significant sentiment features and context elements. +e suggested model gave 93.5% accuracy in the experimental analysis, which was found to be higher than the NB, SVM, and CNN models. Hyun et al. [18] suggested a convolutional neural network model based on target dependence. +e recommended method helps in assessing the impact of the surrounding words on the target word by computing the distance between the target word and the surrounding words. +eir study found that each term in a sentence had a varied effect on the statement's emotional polarity.
A hybrid deep learning model that systematically integrates multiple word embedding approaches (Word2vec, FastText, and character-level embedding) and several deep learning methods (LSTM, GRU, BiLSTM, and CNN) was proposed by researchers in [19]. +e suggested model obtains features by extracting them by using various word embedding methods, merges them, and classifies text as per sentiment. 

To validate the suggested model's performance, numerous deep learning models known as standard models were built and used to run a series of experiments. When comparing the performance of the proposed model with that of earlier research, the new model outperforms the baseline models, according to the findings of this study.

For more information:1950477648nn@gmail.com