International Journal of Knowledge Processing Studies

Abstract While online review platforms are a significant factor in e-commerce decision-making, deceptive reviews can diminish the trustworthiness of review-based information and recommendations. In this research, an empirical model for detecting fake reviews and recommending trustworthy e-commerce is proposed based on traditional machine learning methods. The Deceptive Opinion Spam Corpus consists of 1600 hotel reviews, 800 of which are truthful and 800 of which are deceptive. The text is lowercased, unnecessary punctuation and special characters are removed, the text is tokenized, and the stop words are removed. Textual representation is done using TF-IDF with unigram and bigram features. The classifiers are trained on an 80:20 stratified split of the data with five-fold cross validation on the training set, and the resulting models are compared. The classifiers are trained using an 80:20 stratified split of the data, with five-fold cross validation on the training set, and the resulting models are compared. Models are assessed by accuracy, precision, recall, F1 score and ROC-AUC. SVM achieves the best overall performance, with 89.38% accuracy, 93.13% recall, 89.76% F1-score, and 0.9436 ROC-AUC. The deceptive probability predicted by it is transformed into a review trust score by the formula Trust(r) = 1 − P_fake(r). The review-level trust scores are then aggregated at the hotel level to facilitate the entity ranking based on trust, and the review polarity is provided as supporting information. The framework illustrates how the detection of fake reviews can be used to deliver a credibility signal for more trustworthy recommendation-oriented ranking.

Abstract Though academic libraries have access to wide-ranging circulation data‚ collection decisions are still typically based on descriptive use reports‚ which have shown an inability to identify sustained demand titles‚ or anticipate demand for titles․ A data-driven approach to create ordered demand tiers‚ and an ordered acquisition priorities list‚ from standard circulation data is proposed and validated․ The circulation data for the period June 2024 to June 2026 was collected from the Knowledge Resource Centre‚ SNDT Women's University‚ Pune․ A total of 4984 transactions on 2615 titles and 915 borrowers were analyzed․ Five title-level circulation indicators were employed: total circulation transactions‚ unique borrowers‚ active months‚ transaction per active month‚ and renewals․ The consecutive design involved computing features on the date range of January to December 2025 based on 1‚626 titles‚ and testing prediction on January to April 2026 circulation levels․ We compared K-Means clustering with Ward hierarchical clustering and DBSCAN using four internal indices and a stability test based on 30 random initialisations․ A composite demand score was calculated for the indicators after normalization using four weighting schemes․ However‚ performance of the composite score was similar to a simple transaction-count baseline․ Therefore‚ the main benefit of the composite score was its use of an interpretable tier structure‚ rather than improved ranking performance․ It provides a framework for reproducible acquisition prioritization‚ duplicate-copy planning‚ and identifying underused holdings․