TRANSFORMING PRODUCT INNOVATION TO MEET CUSTOMER NEEDS THROUGH AI MARKETING, A CUSTOMER FEEDBACK ANALYSIS WITH GPT-4O MINI

Sentiment analysis is one part of natural language processing. Sentiment analysis can be done by lexicon based, or machine learning based. Sentiment analysis based on machine learning has advantage of dynamism to meet with new language datasets or new vocabulary. Sentiment analysis seeks to understand the sentiments contained in a sentence. A sentence can be positive, neutral or negative, based on its sentiments. A sentence can have positive, neutral or negative sentiments. However, the fact is each sentence does not always have positive, negative or neutral sentiment clearly. We try to develop a sentiment analysis method that can show the sentiment degree of a sentence. Fuzzy sentiment analysis using convolutional neural network are introduced in this paper to produce more accurate sentiment analysis results. Convolutional neural networks are a popular machine learning method for sentiment analysis. The concept of fuzzy sets is used to express the sentiment degree of a sentence. Euclidean distance analysis to determine the proximity of two vectors is used to show that this method is better than the standard method. The method we propose successfully produces a value that indicates the degree of sentiment of a sentence. Comparison of the euclid distance between the results of the standard sentiment analysis and our method shows that the results of the fuzzy sentiment analysis using convolutional neural network have a distance that is relatively close to the true sentiment value. Fuzzy convolutional neural network analysis sentiment is proven to be able to produce better and smoother sentiment analysis results than standard methods.

Purpose There is a strong need for companies to monitor customer-generated content of social media, not only about themselves but also about competitors, to deal with competition and to assess competitive environment of the business. The purpose of this paper is to help companies with social media competitive analysis and transformation of social media data into knowledge creation for decision-makers, specifically for app-based food delivery companies. Design/methodology/approach Three online app-based food delivery companies, i.e. Swiggy, Zomato and UberEats, were considered in this study. Twitter was used as the data collection platform where customer’s tweets related to all three companies are fetched using R-Studio and Lexicon-based sentiment analysis method is applied on the tweets fetched for the companies. A descriptive analytical method is used to compute the score of different sentiments. A negative and positive sentiment word list is created to match the word present on the tweets and based on the matching positive, negative and neutral sentiments score are decided. The sentiment analysis is a best method to analyze consumer’s text sentiment. Lexicon-based sentiment classification is always preferable than machine learning or other model because it gives flexibility to make your own sentiment dictionary to classify emotions. To perform tweets sentiment analysis, lexicon-based classification method and text mining were performed on R-Studio platform. Findings Results suggest that Zomato (26% positive sentiments) has received more positive sentiments as compared to the other two companies (25% positive sentiments for Swiggy and 24% positive sentiments for UberEats). Negative sentiments for the Zomato was also low (12% negative sentiments) compared to Swiggy and UberEats (13% negative sentiments for both). Further, based on negative sentiments concerning all the three food delivery companies, tweets were analyzed and recommendations for business provided. Research limitations/implications The results of this study reveal the value of social media competitive analysis and show the power of text mining and sentiment analysis in extracting business value and competitive advantage. Suggestions, business and research implications are also provided to help companies in developing a social media competitive analysis strategy. Originality/value Twitter analysis of food-based companies has been performed.

This work focuses on fine-grained sentiment analysis of Arabic text using recent Natural Language Processing methods. Arabic is a language rich in variation, spoken by over 400 million people, yet there is a significant lack of resources for sentiment analysis. To address these challenges, this study employs AraBERT, a model specifically fine-tuned for Arabic text. A corpus of one hundred thousand Arabic reviews across categories such as hotels, books, and movies was scraped and cleaned. These reviews were then categorized into positive, negative, and mixed sentiments. AraBERT was compared with traditional machine learning methods, including Logistic Regression, Decision Tree, Naïve Bayes, and Random Forest. AraBERT achieved superior accuracy of 88\%, along with higher precision, recall, and F1 scores for both positive and negative sentiment classes compared to the other models. This work demonstrates that AraBERT effectively analyzes the syntactic and semantic structure of Arabic, making it a valuable tool for Arabic sentiment analysis across various applications. Future work will extend the model to handle neutral sentiments and include additional dialects to further improve its performance.

At present, Big Data have been created lot of buzz in the technology world. Sentiment Analysis or opinion mining is one of the important applications of 'Big Data', where sentiment analysis is used for recognising voice or response of crowd for products, services. This concept describes the items in some detail and evaluate them as good/bad, preferred/not preferred. The results are very important for a company because customer feedback can yield extremely valuable insights about a company's customer. However, in a commercial website of product reviews, many customers can access to describe the items in some detail and evaluate them with different languages. Therefore, many companies will gather customer feedback in multiple languages. Definitely, feedback in multiple languages raises problems in analysing the material. As this, this paper proposes a solution to classify a product review dataset into two classes: positive and negative sentiments. The proposed methodology is called Multilingual Sentiment Classification (MSC). It consists of two main processing steps: lingual separation and sentiment classification. The first main processing step is to classify online product reviews into language classes. The second processing step is to classify each textual dataset into two classes: positive and negative sentiments. It is noted, we concentrate and experiment on bilingual texts (Thai and English).

Prediction of customer demand is an important part of Supply Chain Management, as it helps to avoid over or under production and reduces delivery time. In the context of e-commerce, accurate prediction of customer demand, typically captured by sales volume, requires careful analysis of multiple factors, namely, type of product, country of purchase, price, discount rate, free delivery option, online review sentiment etc., and their interactions. For e-tailers such as, Amazon, this kind of prediction capability is also extremely important in order to manage the supply chain efficiently as well as ensure customer satisfaction. This study investigates the efficacy of various modeling techniques, namely, regression analysis, decision-tree analysis and artificial neural network, for predicting the sales of books at amazon.in, using various relevant factors and their interactions as predictor variables. Sentiment analysis is carried out to measure the polarity of online reviews, which are included as predictors in these models. The importance of each independent predictor variable, such as discount rate, review sentiment etc., is analyzed based on the outcome of each model to determine top significant predictors which can be controlled by the marketer to influence sales. In terms of accuracy of prediction, the artificial neural network model is found to perform better than the decision-tree based model. In addition, the regression analysis, with and without sentiment and interaction factors, generates comparable results. The comparative analysis of these models reveals several significant findings. Firstly, all three models confirm that review volume is the most important and significant predictor of sales of books at amazon.in. Secondly, discount rate, discount amount and average ratings have minimal or insignificant effect on sales prediction. Thirdly, both negative sentiment and positive sentiment of the reviews are individually significant predictors as per regression and decision-tree model, but they are not significant at all as per neural network model. This observation from the neural network model is contrary to the extant research which claims that both negative and positive sentiment are significant with the former having more influence in predicting sales. Finally, the interaction effects of review volume with negative and positive sentiment are also found to be significant predictors as per all three models. Hence, overall, out of various factors used for sales prediction of books, review volume, negative sentiment, positive sentiment and their interactions are found to be the most significant ones across all models. The results of this study can be utilized by online sellers to accurately predict the sales volume by adjusting these significant factors, thereby managing the supply chain effectively.

PurposeThis study investigates the relationship between earnings call sentiment and subsequent media coverage sentiment. Examining these synergistic effects between executive communication style and resulting news narratives provides novel insights. The unscripted qualitative discussions in earnings calls establish perceptions and outlooks that the media echoes in later coverage. Understanding these intricate connections between information channels aids communication experts and market analysts in shaping strategic messaging and predicting market impacts. In addition, the link with the stock return reaction is revisited, and this study shows that the effects on stock returns driven by news information are moderated by earnings call sentiments.Design/methodology/approachThis study analyzes the interplay between earnings call sentiments and subsequent news sentiments for 30 S&P 500 companies from 2012 to 2022. Utilizing the FinBERT Natural Language Processing (NLP) model, we extract sentiment scores from earnings call transcripts and corresponding news articles. We apply OLS regression models to examine the relationship between negative earnings call sentiments and subsequent negative news sentiments, as well as their combined impact on stock returns. Control variables include financial metrics such as ROA, ROE, firm size, Market-to-Book ratio and liquidity. The methodology allows for a nuanced exploration of sentiment transfer mechanisms in financial communication and their market implications.FindingsOur study reveals a significant positive correlation between negative sentiment in earnings calls and subsequent negative news sentiment. A 1% increase in negative call sentiment associates with a 0.54% increase in negative news sentiment the following day, supporting Agenda Building and Impression Management hypotheses. We observe a multiplicative effect on stock returns when negative call sentiment coincides with negative news sentiment, supporting signaling theory. Financial metrics like ROE show marginal influence on news sentiment, while others demonstrate insignificant impact. These findings underscore the importance of holistic corporate communication management in mitigating potential negative market reactions.Research limitations/implicationsThis study’s primary limitation is its sample size of 30 S&P 500 companies, potentially limiting generalizability. The use of a single sentiment analysis model (FinBERT) could impact results, warranting comparison with alternative methods. The study’s timeframe (2012–2022) may not capture the most recent market dynamics. Future research could expand the sample size, incorporate additional sentiment analysis techniques and explore longer-term effects. Investigating industry-specific variations and the impact of macroeconomic factors could provide further insights. Additionally, qualitative analysis of earnings call content could complement these quantitative findings, offering a more comprehensive understanding of sentiment transfer mechanisms.Practical implicationsThis study offers insights for corporate communicators, investor relations professionals and financial analysts. The strong correlation between earnings call sentiment and subsequent news sentiment emphasizes the need for management of corporate messaging during these calls. Companies should be aware that negative sentiments expressed in earnings calls may amplify through news coverage, potentially impacting stock performance. Investors and analysts should consider both earnings call and news sentiments when evaluating market reactions. For regulators, these findings highlight the importance of monitoring information dissemination practices to ensure market fairness. Overall, the study underscores the significance of a holistic approach to financial communication strategy.Social implicationsThis research highlights the interconnected nature of corporate communication and media narratives, emphasizing social responsibility of both corporations and news outlets. The findings suggest that negative corporate messaging can perpetuate and amplify through news coverage, potentially affecting public perception and investor sentiment. This underscores the need for transparent and ethical communication practices in the business world. The study also raises awareness about the potential manipulation of public opinion through carefully crafted corporate narratives. It encourages stakeholders to critically evaluate both corporate communications and subsequent media coverage, promoting a more informed and discerning society in the context of financial information dissemination.Originality/valueThis study uniquely explores the interplay between earnings call sentiments and subsequent news sentiments, addressing a significant gap in financial communication research. By examining the sentiment transfer mechanism from corporate messaging to media narratives, it provides novel insights into information dissemination in financial markets. The research demonstrates how negative sentiments in earnings calls can amplify through news coverage, offering valuable implications for corporate communication strategies. This multifaceted analysis contributes to a deeper understanding of the complex relationships between corporate communication, media coverage and market behavior.

Sentiment analysis is a natural language processing (NLP) technique for determining emotional tone in a body of text. Using product reviews in sentiment analysis and opinion mining various methods have been developed previously. Although, existing product review analyzing techniques could not accurately detect the product aspect and non-aspect. Hence a novel Detach Frequency Assort is proposed to detect the product aspect term using TF-ISF (Term frequency-inverse sentence frequency) with Part of Speech (POS) tags for sentence segmentation and additionally using Feedback Neural Network to combine product aspect feedback loop. Furthermore, decision-making problem occurs during classification of sentiments. Hence, to solve this problem a novel technique named, Systemize Polarity Shift is proposed in which flow search based Support Vector Machine (SVM) with Bag of Words model classifies pre-trained review comments as positive, negative, and neutral sentiments. Moreover, the identification of specific products is not focused in sentiment analysis. Hence, a novel Revival Extraction is proposed in which a specific product is extracted based on thematic analysis method to obtain accurate data. Thus, the proposed Product Review Opinion framework gives effective optimized results in sentiment analysis with high accuracy, specificity, recall, sensitivity, F1-Score, and precision.

This study compares the effectiveness of the Support Vector Machine (SVM) and Naïve Bayes algorithms in classifying user sentiment regarding the BRImo application. User reviews were obtained from the Google Play Store platform and underwent a text preprocessing stage to clean and prepare the data. Subsequently, the SVM and Naïve Bayes algorithms were applied for sentiment analysis, using evaluation metrics such as accuracy, precision, recall, and F1-score. The results show that SVM achieved a training accuracy of 95.67% and a testing accuracy of 83.11%, with its best performance on positive sentiment (precision 92.26%, recall 91.79%, F1-score 92.02%) and moderate performance on negative sentiment (precision 62.81%, recall 62.81%, F1-score 62.81%). Meanwhile, Naïve Bayes recorded a training accuracy of 95.23% and a testing accuracy of 82.77%, with its highest performance on positive sentiment (precision 90.12%, recall 93.38%, F1-score 91.72%) but lower performance on negative sentiment (precision 65.07%, recall 60.06%, F1-score 62.46%). In terms of sentiment distribution, SVM was more effective in handling sentiment variations, particularly in detecting negative and neutral sentiments. These findings indicate that SVM outperforms Naïve Bayes in sentiment analysis of user reviews for the BRImo application.

The presence and significance of artificial intelligence (AI) technology in society have been steadily increasing since 2000. While its potential benefits are widely acknowledged, concerns about its impact on society, the economy, and ethics have also been raised. Consequently, artificial intelligence has garnered widespread attention in news media and popular culture. As mass media plays a pivotal role in shaping public perception, it is crucial to evaluate opinions expressed in these outlets. Understanding the public’s perception of artificial intelligence is essential for effective public policy and decision making. This paper presents the results of a sentiment analysis study conducted on WIRED magazine’s coverage of artificial intelligence between January 2018 and April 2023. The objective of the study is to assess the prevailing opinions towards artificial intelligence in articles from WIRED magazine, which is widely recognized as one of the most reputable and influential publications in the field of technology and innovation. Using two sentiment analysis techniques, AFINN and VADER, a total of 4265 articles were analyzed for positive, negative, and neutral sentiments. Additionally, a term frequency analysis was conducted to categorize articles based on the frequency of mentions of artificial intelligence. Finally, a linear regression analysis of the mean positive and negative sentiments was performed to examine trends for each month over a five-year period. The results revealed a leading pattern: there was a predominant positive sentiment with an upward trend in both positive and negative sentiments. This polarization of sentiment suggests a shift towards more extreme positions, which should influence public policy and decision making in the near future.

BackgroundSocial media serves as a vast repository of data, offering insights into public perceptions and emotions surrounding significant societal issues. Amid the COVID-19 pandemic, long COVID (formally known as post–COVID-19 condition) has emerged as a chronic health condition, profoundly impacting numerous lives and livelihoods. Given the dynamic nature of long COVID and our evolving understanding of it, effectively capturing people’s sentiments and perceptions through social media becomes increasingly crucial. By harnessing the wealth of data available on social platforms, we can better track the evolving narrative surrounding long COVID and the collective efforts to address this pressing issue.ObjectiveThis study aimed to investigate people’s perceptions and sentiments around long COVID in Canada, the United States, and Europe, by analyzing English-language tweets from these regions using advanced topic modeling and sentiment analysis techniques. Understanding regional differences in public discourse can inform tailored public health strategies.MethodsWe analyzed long COVID–related tweets from 2021. Contextualized topic modeling was used to capture word meanings in context, providing coherent and semantically meaningful topics. Sentiment analysis was conducted in a zero-shot manner using Llama 2, a large language model, to classify tweets into positive, negative, or neutral sentiments. The results were interpreted in collaboration with public health experts, comparing the timelines of topics discussed across the 3 regions. This dual approach enabled a comprehensive understanding of the public discourse surrounding long COVID. We used metrics such as normalized pointwise mutual information for coherence and topic diversity for diversity to ensure robust topic modeling results.ResultsTopic modeling identified five main topics: (1) long COVID in people including children in the context of vaccination, (2) duration and suffering associated with long COVID, (3) persistent symptoms of long COVID, (4) the need for research on long COVID treatment, and (5) measuring long COVID symptoms. Significant concern was noted across all regions about the duration and suffering associated with long COVID, along with consistent discussions on persistent symptoms and calls for more research and better treatments. In particular, the topic of persistent symptoms was highly prevalent, reflecting ongoing challenges faced by individuals with long COVID. Sentiment analysis showed a mix of positive and negative sentiments, fluctuating with significant events and news related to long COVID.ConclusionsOur study combines natural language processing techniques, including contextualized topic modeling and sentiment analysis, along with domain expert input, to provide detailed insights into public health monitoring and intervention. These findings highlight the importance of tracking public discourse on long COVID to inform public health strategies, address misinformation, and provide support to affected individuals. The use of social media analysis in understanding public health issues is underscored, emphasizing the role of emerging technologies in enhancing public health responses.

BackgroundSocial media platforms have become influential spaces for disseminating information about electronic cigarettes (e-cigarettes). Concerns persist about the spread of misleading content, particularly among social media vulnerable groups. Xiaohongshu (RedNote), widely used by Chinese youth, plays a growing role in shaping e-cigarette perceptions. Understanding the narratives circulating on this platform is essential for identifying misinformation, assessing public perception, and guiding future health communication strategies.ObjectiveThis study aimed to analyze the content, topics, user engagement, and sentiment trends of e-cigarette–related posts on Xiaohongshu and to assess the factors that influence engagement.MethodsE-cigarette–related posts published on Xiaohongshu between January 2020 and November 2024 were collected using web scraping, based on a predefined keyword list and a time-stratified random sampling strategy. Posts were categorized into 4 themes: advertising promotion, health hazards, usage interaction, and others. High-frequency keywords were extracted, and representative quotes were included to illustrate user perspectives across each category. Sentiment analysis was performed on posts in the usage interaction category to assess public attitudes. We defined 4 sentiment categories: positive, negative, neutral, and mixed. Logistic regression was conducted to explore the effects of post type, content length, and thematic classification on user engagement metrics such as likes, saves, and comments.ResultsA total of 1729 posts were included and analyzed. Usage interaction posts were the most common (681/1729, 39.39%), with keywords such as “experience,” “regulations,” and “quit smoking” dominating this category. Advertising promotion posts (512/1729, 29.61%) frequently used terms like “flavor,” “fashion,” and “design” to attract younger users. Health hazards posts (311/1729, 17.99%) highlighted risks with keywords like “nicotine,” “addiction,” and “secondhand smoke,” while others included policy and industry updates. Representative quotes highlighted typical concerns about aesthetics, health risks, and cessation struggles. Health hazards posts garnered the highest engagement in terms of likes and saves, despite their limited presence (odds ratio [OR] 1.498, 95% CI 1.099‐2.042, P=.01). Video posts significantly outperformed text-image posts in generating comments (OR 2.624, 95% CI 2.017‐3.439, P<.001). Sentiment analysis of the usage interaction posts (n=681) revealed that 53.45% (364/681) were positive, highlighting reduced harm, convenience, or flavor preferences. Negative sentiment was observed in 33.48% (228/681) of posts, often expressing concerns about addiction and health risks. Mixed sentiments appeared in 6.90% (47/681), acknowledging both pros and cons. In addition, 6.17% (42/681) of posts were classified as neutral without evident emotional tone.ConclusionsThe findings underscore the dual role of Xiaohongshu as a platform for both e-cigarette promotion and public discourse. Misleading marketing targeting vulnerable groups, such as adolescents, remains a critical issue. However, the strong user response to health-related content suggests that social media platforms could be leveraged for effective health education. Strengthened regulatory oversight and educational campaigns leveraging engaging content formats are urgently needed to counter misinformation and protect public health.

Sentiment analysis is a classification procedure where we apply machine learning and deep learning algorithms to analyze the sentiment of the dataset, which consists of text, e.g., a message that can be of positive or negative sentiment. In this study, an attempt has been made to investigate which sentiment analysis techniques are feasible for product reviews. Here, the Amazon reviews dataset is used to compare, train, and test various machine learning and deep learning methods having product reviews from Amazon, which are chosen randomly from an open-source repository. The dataset comprises 4 million reviews. Comparison of several algorithms' performances, i.e., RFC, XGBC, LGBM, MNB, GBC, DTC, and Bi-LSTM, amongst which Bi-LSTM gives the highest performance among the algorithms used for classification. It was also applied to the other reviews from the Amazon dataset to predict the sentiment of the reviews, as well as a fresh Amazon scraped dataset comprising product reviews from several categories. This resulted in a very accurate classification, with the best results for test reviews on the amazon dataset. In conclusion, Bi-LSTM networks are excellent for categorizing customer sentiment on product reviews, and the results do not differ considerably across categories.

Organisations these days are actively using social media platforms to engage with potential and existing customers and monitor what they say about the organisation’s product or service. The most important area within social media monitoring lies in how to gain insight for sentiment analysis. Sentiment analysis helps in effective evaluation of customer’s sentiments in real time and takes on a special meaning in the context of online social networks like Twitter and Facebook, which collectively represent the largest online forum available for public opinion. Sentiment Analysis is not about retrieving and analyzing the analytics purely on the basis of positive, negative or neutral sentiment. It is imperative to assess the influencers of the sentiments in terms of Retweet and Share option used by them on Twitter and Facebook platform respectively. Measuring the intensity is other important aspect of sentiment analysis process. What kind of nouns, adjectives, verbs and adverbs are used in the opinion across the Twitter and Facebook platform matters as well since it exhibits the intensity of the underlying emotion in the text written. This study was conducted to propose a framework to identify and analyse the positive and negative sentiments present in Twitter and Facebook platforms and an algorithm was prepared to measure the intensity and influence of the positive, negative sentiment in particular using the document and sentence level analysis technique.

The waste problem is a severe problem that significantly affects the environment and public health. To effectively determine the public’s perception of the waste problem, it is necessary to examine public sentiment toward waste management. This research aims to develop a sentiment analysis model using VADER and deep-translator and analyze the Yogyakarta waste emergency problem. This research was conducted in two phases, namely, the first phase was developing a sentiment analysis model by evaluating its performance based on public data. Then, the second phase classifies public comments from YouTube regarding the waste problem to understand public perceptions and evaluations by identifying positive, negative, and neutral sentiments. The model evaluation results show that sentiment analysis using VADER and deep translator can achieve Accuracy, Precision, Recall, and F1-score values of 0.716, 0.837, 0.853, and 0.738, respectively. The sentiment results from YouTube comments obtained positive, neutral, and negative sentiments of 30.0%, 31.7%, and 37.3%, respectively. The results of the sentiment analysis are neutral sentiment discussing waste management, disappointment in negative sentiment, and hope for waste management in positive sentiment.

Multifocal intraocular lens implants (IOLs) are a premium option for cataract surgery which patients may purchase to achieve improved spectacle-independence for near vision but may have trade-offs with visual quality. We demonstrate the use of sentiment analysis to evaluate multifocal lenses discussed on MedHelp, a leading online health forum. A search for "multifocal IOL" was performed on MedHelp.org on November 1, 2016, yielding relevant patient posts. Sentiment analysis was performed using IBM's Watson, which extracted 30,066 unique keywords and their associated sentiment scores from 7495 posts written by 1474 unique patient users. Keywords associated with monovision, monofocal, and toric lenses had positive mean sentiment, significantly higher than for keywords associated with multifocals, which had negative mean sentiment (p < 0.001, ANOVA). Many keywords represented complaints and were associated with negative sentiment, including glare, halo, and ghosting. Sentiment analysis can provide insights into patient perspectives towards multifocal lenses by interpreting online patient posts.