Advanced Information Retrieval System: Theoretical and Experimental Perspective

The vast data sources use Information Retrieval (IR) methods for accessing the required information. The classical methods of IR, like Boolean retrieval and vector space, are fundamental for search strategies. The integration of these techniques with artificial intelligence will improve the accuracy of information retrieval. The research work compares the working of both the classical and modern information retrieval techniques. The comparison provides the merits and demerits of both methods so that information retrieval can be made more intelligent. The classical methods include Boolean retrieval and Term Frequency - Inverse Document Frequency (TF-IDF), while the modern method is Bidirectional Encoder Representations from Transformers (BERT). The authors have used the MS MARCO dataset for training and testing purposes. The authors have calculated precision, recall, and F1 score to evaluate the research work. From the simulation of work, it has been noticed that modern information retrieval methods can reduce the computational cost and enhance the accuracy of retrieval.

Information Retrieval (IR) techniques are growing continuously from being keyword-based systems to advanced search. These days, IR techniques utilize Machine Learning (ML), Deep Learning (DL), and Natural Language Processing (NLP) for providing more accurate and personalized results. In the proposed research work, the IR techniques are analysed for their merits and demerits. In the work, it has been examined how contemporary research has been transformed into query document matching. This work integrates Term Frequency-Inverse Document Frequency (TFIDF) into two retrieval metrics—cosine similarity and dot product similarity. Integration aims to provide better results. Cosine similarity is good at capturing vector orientation, while dot product similarity is good for vector magnitude. A combined similarity is weighted at parameter α to enhance the retrieval capacity. From the simulation of work, it has been calculated that the combined method performed well. In the future, authors will incorporate machine learning or deep learning methods to enhance the performance of these IR techniques.

The significance of an improved user experience in recommender systems will continue to grow as tourism advances. Most such systems generally offer information to users that benefits third-party firms' commercial interests, often involving the implicit compromise of user privacy. In the proposed work, a tourism recommendation system is developed by the authors to deliver personal recommendations to users. For designing the proposed recommendation system, the authors have considered parameters such as ratings, user activity, time utilized by the user, certifications, and other relevant attributes. Four methods: Memory-based Collaborative Filtering (MCF), Matrix factorization Collaborative Filtering (MFCF), Cosine Similarity Content Filtering (CSCF), and Term Frequency-Inverse Document Frequency (TF-IDF) are used by the authors for designing a tourist recommendation system. For the assessment of the work, the authors have calculated Mean Average Error (MAE) and Root Mean Square Error (RMSE). 

The user engagement in YouTube comments could be utilized for sentiment analysis. The data is collected through web scraping from the YouTube comments. Natural Language Processing (NLP) and Bidirectional Encoder Representations from Transformers (BERT) are utilized by authors to analyze the sentiment of commenters. The current trends and user engagement in the market can be analyzed through YouTube comments. The thematic patterns are generated with the help of word clouds and sentiment distribution charts. Based on the outcome of the proposed work, it can be concluded that YouTube comments provide valuable feedback to researchers, which in turn can be used for sentiment analysis. 

With the continuous growth of user-generated data on social media platforms, its analysis requires a deep understanding of sentiment trends. Data from social sites is collected using web scraping techniques and then filtered for consistency through pre-processing methods. In this study, multiple techniques are applied to analyze YouTube comments. The BERT transformer model is used to classify comments into positive, negative, and neutral categories, enabling sentiment analysis and providing insights into user opinions and trends. Additionally, Latent Dirichlet Allocation (LDA) is employed for thematic analysis to identify key discussion topics within the comments. The performance of the proposed approach is evaluated using the F1-score metric. For future improvements, deep learning techniques could be explored to enhance the accuracy of sentiment analysis.

In the digital era, sentence interpretation is crucial for understanding the meaning of sentences. As a key component of Natural Language Processing (NLP), it helps identify relationships between words and determine their roles within a sentence. Semantic Role Classification (SRC) assigns semantic roles to different actions in a sentence, enabling deeper language comprehension. This study analyses various SRC techniques, with a particular focus on transformer models. It provides a summary of existing SRC methods, highlighting their advantages and incorporating a Continuous Integration/Continuous Delivery (CI/CD) pipeline for seamless deployment. The effectiveness of the proposed approach is evaluated based on the accuracy achieved.

This chapter presents the classification and analysis of fashion data, which consists of 90 images belonging to one class, using deep learning techniques. Data augmentation is done to pre-process the dataset. Features are retrieved using Convolutional Neural Networks (CNNs), VGG16, and ResNet50. These modes are trained on styles and patterns of images so that recognition can be done. For the styles and subtitles, another dataset of 144 audio files has been utilized. Voice is converted into text by using Machine Learning (ML) and Natural Language Processing (NLP) techniques. Pre-processing of audio files has been performed using Mel-Frequency Cepstral Coefficients (MFCC) along with normalization to reduce noise. The Recurrent Neural Networks (RNNs) technique converts the audio file into a text file. The proposed work is evaluated based on accuracy, reliability, and adaptability. 

Due to the digital era, there is a flood of information available online . Recommendation systems can be utilized to improve user experience. Authors in this chapter proposed to develop a hybrid book recommendation system using Collaborative Filtering (CF) and Content-based Filtering (CBF) techniques. The content filtering method identifies the features of the object, and then, based on those features, a decision can be made. Collaborative filtering works based on user interaction patterns. The proposed method overcomes the drawbacks of data sparsity in collaborative filtering and the cold-start problem in content-based systems. To overcome these issues, the dataset is pre-processed to find the popular books, ratings, and a list of active users. Authors have applied k-Nearest Neighbors (k-NN) for feature and metric selection so that the recommendation engine can work properly. The performance of the proposed model is measured in terms of performance metrics, and the outcome has proved that it is a precise and accurate recommendations model with high recall, precision, F1-Score, and accuracy metrics. 

 Due to advancements in computer vision techniques, medical healthcare data is available to users in a very large amount. With the use of this data, an intelligent medicine recommendation system can be prepared. Authors in this chapter utilized Term Frequency-Inverse Document Frequency (TF-IDF) and Machine Learning (ML) for the development of an intelligent medicine recommendation system. The recommendation system recommends proper medicines by exploring the symptoms of patients and their medical history. The TF-IDF technique retrieves suitable features from the dataset and then applies machine learning for the classification of diseases and the recommendation of medicine to patients. The proposed recommendation system provides valuable and accurate suggestions to users, adding flexibility to the healthcare system.

The fashion industry has undergone a transformation thanks to the rapid development of computer vision, which has enabled the automatic classification and analysis of clothing products. In this work, Convolutional Neural Networks (CNNs) are used to categorize a varied fashion dataset with more than 15,000 images from Kaggle. The dataset's diversity of clothes and accessory categories reflects the intricacy of realworld fashion datasets. The authors focused on data preprocessing, augmentation, and fine-tuning to enhance the model's classification accuracy using a CNN. The aim of this work is to demonstrate how well CNNs handle visual diversity and identify patterns in the dataset while addressing issues such as dataset imbalance and inter-class similarity.

For validation of the proposed work, the authors have calculated accuracy, precision, recall, and F1-score. The outcome of the CNN model demonstrates that Artificial Intelligence (AI) has its application in the fashion industry, and performance metrics justified the work. The confusion metric highlights the merits and demerits of the CNN model. In the future, a bigger dataset could be utilized for the incorporation of transfer learning, resulting in enhanced accuracy.

The rapid expansion of scientific literature has made it challenging for researchers to find relevant studies on specific topics efficiently. The traditional search methods require extensive time and manual effort to identify, filter, and extract essential information from numerous sources. This paper proposes an automated keyword-based web crawling system aimed at streamlining the retrieval of research papers and patents from sources, IEEE and Google patents, respectively. In the proposed work, the authors type the keyword of the research paper or patent, and then the system provides the results accordingly. The details of a patent or research paper may include details like the name of the author, the DOI, the publisher, the title of the article, publication date, and many more. The proposed technique decreases manual work and enhances the accuracy of collecting the required data.