Clustering short temporal behaviour sequences for customer segmentation using LDA

Problem: This work systematically investigates the effectiveness of various visual feature coding schemes for facilitating the learning of timedelayed dependencies among disjoint multi-camera views. Related work: Quite a few studies [3, 4, 6] have been proposed to model inter-camera dependency across non-overlapping camera views. Learning time-delayed correlations among disjoint cameras in crowded public scenarios is a non-trivial task: (1) the time gaps between camera views are unknown therefore activities in two related views may occur at arbitrary time delays with high uncertainty; (2) the features are inevitably noisy, ambiguous, and may vary drastically across views owning to illumination condition, camera angles, and changes in object pose. Most state-ofthe-art methods typically hand pick a few features tailored to the target environment, with the hope that those chosen features contain robust and sufficient statistics for correlating the time-delayed activity patterns across disjoint views. These manual approaches to hard selection of features are neither principled nor generalisable to different scene context. Our solution: In this study, we wish to examine the concept that visual features should be coded and selected automatically for robust and accurate time-delayed dependency learning. The contributions of this study are two-fold: (1) We present a systematic study and evaluation to investigate the effectiveness of supervised and unsupervised feature coding methods to facilitate the learning of inter-camera activity pattern dependencies. (2) We systematically evaluate the sensitivity of inter-camera time delayed dependency learning given different training video sizes and region decomposition qualities. These factors are critical for accurate dependency learning but have been largely ignored by the published existing work in the literature. Approach overview: We employ the Random Forest [2] as the supervised feature coding approach. In particular, given a set of localised features extracted from a region, together with people count training label over time, we first train a regression forest to learn the non-linear mapping between the crowd density and the corresponding low-level features. Given unseen data, we then construct a time series based on the predicted crowd density ŷ obtained from the regression forest (RF pred), the treestructured code (tree code) [5], or the combination of the two. As for unsupervised coding scheme, we use the Latent Dirichlet Allocation (LDA) [1] to map the low-level features into codewords that capture the topic distribution, whereby an image region patch (document) d is treated as a collection of j = 1 . . .Ni features (words). To form the unsupervised feature codes, given a sequence of localised feature vectors detected from a region, we first perform quantisation on each feature to generate a bag-of-word representation for all image patches. Similar to text documents, these bag-of-word represented image patches are fed into the LDA, which gives us a topic-based representation. Once having the topic-based code (topic code), we perform k-means quantisation on them, producing the final compact topic-based code, and concatenate them over time to form a time series. To solve the problem of using the feature codes for learning intercamera dependencies, we adopt the Time Delayed Mutual Information (TDMI) proposed in [3] due to its reported effectiveness and simplicity. The input to TDMI are time series generated from either the supervised or the unsupervised coding scheme. In addition to measuring deviation error in transition time, we propose a new metrics to evaluate the effectiveness of different coding methods, called Mutual Information Margin (MIM):

This study conducts a descriptive bibliometric analysis of artificial intelligence (AI) applications in banking systems, synthesizing 622 peer-reviewed journal articles published between 1996 and 2024. Drawing from the Web of Science Core Collection, the dataset was screened to include only ABS-listed or Scopus-indexed journals. The analysis applies keyword co-occurrence, citation profiling, Latent Dirichlet Allocation (LDA), and SCOR-style process classification to identify thematic clusters and research gaps. Findings show that scholarly output has increased significantly since 2019, with machine learning and predictive models dominating the methodological landscape. Most studies focus on credit scoring and fraud detection, while compliance, investment advisory, and prescriptive analytics remain marginally addressed. Five thematic research clusters were identified: model evaluation, fintech integration, credit classification, organizational transformation, and decision support. Journals such as the International Journal of Bank Marketing and Annals of Operations Research were among the most prolific sources. Despite progress, the literature remains imbalanced favouring technical outputs over behavioural, ethical, or institutional dimensions. This paper offers a structured research agenda emphasizing decision-oriented AI models, compliance analytics, human-AI collaboration, and strategic integration. The results inform scholars and banking professionals seeking to align AI innovations with financial governance, digital transformation, and sustainable operational design.

Specification, estimation, and analysis of macroeconomic models: Ray C. Fair. Cambridge, MA: Harvard University Press, 1984. 479 pp. $35.00 ISBN 0-674-83180-2

In recent past, there is an increased interest in time series clustering research, particularly for finding useful similar trends in multivariate time series in various applied areas such as environmental research, finance, and crime. Clustering multivariate time series has potential for analyzing large volume of crime data at different time points as law enforcement agencies are interested in finding crime trends of various police administration units such as states, districts and police stations so that future occurrences of similar incidents can be overcome. Most of the traditional time series clustering algorithms deals with only univariate time series data and for clustering high dimensional data, it has to be transformed into single dimension using a dimension reduction technique. The conventional time series clustering techniques do not provide desired results for crime data set, since crime data is high dimensional and consists of various crime types with different weight age. In this paper, a novel approach based on dynamic time wrapping and parametric Minkowski model has been proposed to find similar crime trends among various crime sequences of different crime locations and subsequently use this information for future crime trends prediction. Analysis on Indian crime records show that the proposed technique generally outperforms the existing techniques in clustering of such multivariate time series data.

Robust Monitoring of Time Series with Application to Fraud Detection

Because of the growing involvement of communities from various disciplines, data science is constantly evolving and gaining popularity. The growing interest in data science-based services and applications presents numerous challenges for their development. Therefore, data scientists frequently turn to various forums, particularly domain-specific Q&A websites, to solve difficulties. These websites evolve into data science knowledge repositories over time. Analysis of such repositories can provide valuable insights into the applications, topics, trends, and challenges of data science. In this article, we investigated what data scientists are asking by analyzing all posts to date on DSSE, a data science-focused Q&A website. To discover main topics embedded in data science discussions, we used latent Dirichlet allocation (LDA), a probabilistic approach for topic modeling. As a result of this analysis, 18 main topics were identified that demonstrate the current interests and issues in data science. We then examined the topics' popularity and difficulty. In addition, we identified the most commonly used tasks, techniques, and tools in data science. As a result, "Model Training", "Machine Learning", and "Neural Networks" emerged as the most prominent topics. Also, "Data Manipulation", "Coding Errors", and "Tools" were identified as the most viewed (most popular) topics. On the other hand, the most difficult topics were identified as "Time Series", "Computer Vision", and "Recommendation Systems". Our findings have significant implications for many data science stakeholders who are striving to advance data-driven architectures, concepts, tools, and techniques.

Data mining examines large pre-existing databases in order to generate new information. There are various tasks included under Data mining and association rule mining is considered as one of the crucial tasks among its. They are in form of if-then kind of statements which help to find relationships among huge data which do not hold relationship with each other within a relational database or any other information repository. As there are many applications like market basket analysis, detection of fraud in web, medical diagnosis, census data, Customer Relationship Management of credit card business which uses association rules so it is possible to improve the process of Decision making. Security is required for individual transaction and for frequent itemsets when the database is partitioned horizontally among multiple sites. In this case, every site is interested in globally supported association rules without revealing its own local information. To fulfill this goal, We use a secure multi-party algorithm based on secure sum technique to simplify the operation of mining association rule when the database is horizontally partitioned among multiple sites. We are using a Frequent-Pattern (FP) growth algorithm to find frequent itemsets and try to reduce total computation time.

Phishing is an online criminal activity which traps regular online customers to reveal their sensitive information into fake destinations by disguising themselves as original website. Because of the short life expectancy of phishing sites and the fast progression of phishing systems the current anti- phishing solutions are either unfit to manage the rising changes or in fit for assuming a successful part against this attack.Natural language handling and machine learning – these are the centre usage of the multi-stage approach proposed to identify phishing attacks and to perceive the substance/association that has been misused by the attackers to execute the phishing attacks. In this technique the primary stage is the revelation of named elements (names of areas, individuals and associations) and afterward the disclosure of concealed topics and for this the strategies that backings both phishing and non-phishing information therefore, Conditional Random Field (CRF) and Latent Dirichlet Allocation (LDA) is utilized. Next stage is the AdaBoost arrange where the named elements and the concealed subjects are dealt as features and the messages are ordered into phishing or non-phishing. proposed techniques are planned to alleviate the effect of phishing attack. The principal strategy proposes is to identify Wi-Fi phishing utilizing Association Rule mining. A Wi-Fi hotspot which associates such hand held gadgets has turned into a transitory archive of delicate data, in this way giving an open door for programmers to gather money related gain .Consequently data security arrangements tending to Wi-Fi hotspots (Wi-Fi phishing attack) has turned into the need of great importance. The conceivable approaches to attack the security of Wi-Fi hotspots and counter-attack methodologies have been tended to in this strategy. Key works: Phishing, LDA,AI

This paper focuses on two types of modeling of information flow in news stream, namely, burst analysis and topic modeling. First, when one wants to detect a kind of topics that are paid much more attention than usual, it is usually necessary for him/her to carefully watch every article in news stream at every moment. In such a situation, it is well known in the field of time series analysis that Kleinberg’s modeling of bursts is quite effective in detecting burst of keywords. Second, topic models such as LDA (latent Dirichlet allocation) are also quite effective in estimating distribution of topics over a document collection such as articles in news stream. However, Kleinberg’s modeling of bursts is usually applied only to bursts of keywords but not to those of topics. Considering this fact, we propose how to apply Kleinberg’s modeling of bursts to topics estimated by a topic model such as LDA and DTM (dynamic topic model).

BackgroundBioinformatics is an interdisciplinary field at the intersection of molecular biology and computing technology. To characterize the field as convergent domain, researchers have used bibliometrics, augmented with text-mining techniques for content analysis. In previous studies, Latent Dirichlet Allocation (LDA) was the most representative topic modeling technique for identifying topic structure of subject areas. However, as opposed to revealing the topic structure in relation to metadata such as authors, publication date, and journals, LDA only displays the simple topic structure.MethodsIn this paper, we adopt the Tang et al.’s Author-Conference-Topic (ACT) model to study the field of bioinformatics from the perspective of keyphrases, authors, and journals. The ACT model is capable of incorporating the paper, author, and conference into the topic distribution simultaneously. To obtain more meaningful results, we use journals and keyphrases instead of conferences and bag-of-words.. For analysis, we use PubMed to collected forty-six bioinformatics journals from the MEDLINE database. We conducted time series topic analysis over four periods from 1996 to 2015 to further examine the interdisciplinary nature of bioinformatics.ResultsWe analyze the ACT Model results in each period. Additionally, for further integrated analysis, we conduct a time series analysis among the top-ranked keyphrases, journals, and authors according to their frequency. We also examine the patterns in the top journals by simultaneously identifying the topical probability in each period, as well as the top authors and keyphrases. The results indicate that in recent years diversified topics have become more prevalent and convergent topics have become more clearly represented.ConclusionThe results of our analysis implies that overtime the field of bioinformatics becomes more interdisciplinary where there is a steady increase in peripheral fields such as conceptual, mathematical, and system biology. These results are confirmed by integrated analysis of topic distribution as well as top ranked keyphrases, authors, and journals.

After the East Japan Great Earthquake happened on Mar. 11, 2011, many affected people who lost houses, jobs and families fell into difficulties. Governmental agencies and NPOs supported them by offering relief supplies, foods, evacuation centers and temporary houses. When various supports were offered to affected people, if Governmental agencies and NPOs could detect their needs appropriately, it was effective for supporting them. This paper proposes the method to extract affected people's needs from Social Media after the Earthquake and analyze their needs changes over time. We target the blog that expressed thoughts, requirements and complaints of affected people, and adopt the Latent Dirichlet Allocation (LDA) that is one of popular techniques for topic extraction. We then compare the analysis result with affected people's actual situation and real events and evaluate the effectiveness of our method. In addition, we evaluate the effectiveness as the method that can help decision making for providing appropriate supports to affected people.

This article explores the transformative impact of Artificial Intelligence (AI) on subscription-based business models across various industries. It examines how AI is revolutionizing key aspects of subscription services, including personalization, customer retention, pricing strategies, customer support, operational efficiency, and fraud detection. The article highlights specific AI applications such as content recommendations, dynamic user interfaces, churn prediction, advanced customer segmentation, dynamic and usage-based pricing, AI-powered chatbots, and sentiment analysis. Additionally, it discusses how AI enhances operational efficiency through automated billing and inventory management, and improves security via anomaly detection and behavioral biometrics. Case studies of Adobe Creative Cloud and Amazon Web Services (AWS) are presented to illustrate real-world applications of AI in subscription services. The article concludes by emphasizing the paradigm shift AI represents in customer engagement, operational optimization, and revenue generation for subscription businesses, forecasting continued innovation and opportunities in this rapidly evolving sector.

Billing Mediation Platform (BMP) in Telco is used to process real-time streams of Call Detail Records (CDRs) which can number tens of billions a day. The comprehensive records generated by BMPs can be used for billing and accounting, fraud detection, campaign management, spam filtering, traffic analysis, and churn prediction. Many of these applications are characterized by real-time processing requiring high throughput, low-latency analysis of CDRs. Testing such BMPs has different dimensions, stress testing of analytics for scalability, correctness of analytics, what-if scenarios, all of which require CDRs with realistic volumetric and contextual properties. We propose WLG, a framework for testing and benchmarking BMPs which involves generating high volumes of CDRs representative of real-world data. The framework is flexible in its ability to express and tune the workload generation to simulate CDRs from broad range of traffic patterns while preserving different spatio-temporal correlations and content-level information observed in real-world CDRs.

Due to high dimensionality and multiple variables, unsupervised classification of multivariate time series (MTS) involves more challenging problems than those of univariate ones. Unlike the vectorization of a feature matrix in traditional clustering algorithms, an unsupervised pattern recognition scheme based on matrix data is proposed for MTS samples in this paper. To reduce the computational load and time consumption, a novel variable-based principal component analysis (VPCA) is first devised for the dimensionality reduction of MTS samples. Afterward, a spatial weighted matrix distance-based fuzzy clustering (SWMDFC) algorithm is proposed to directly group MTS samples into clusters as well as preserve the structure of the data matrix. The spatial weighted matrix distance (SWMD) integrates the spatial dimensionality difference of elements of data into the distance of MST pairs. In terms of the SWMD, the MTS samples are clustered without vectorization in the dimensionality-reduced feature matrix space. Finally, three open-access datasets are utilized for the validation of the proposed unsupervised classification scheme. The results show that the VPCA can capture more features of MTS data than principal component analysis (PCA) and 2-D PCA. Furthermore, the clustering performance of SWMDFC is superior to that of fuzzy c -means clustering algorithms based on the Euclidean distance or image Euclidean distance.

Classification and associative rule mining are two substantial areas in data mining. Some scientists attempt to integrate these two field called rule-based classifiers. Rule-based classifiers can play a very important role in applications such as fraud detection, medical diagnosis, and etc. Numerous previous studies have shown that this type of classifiers achieves high classification accuracy than traditional classification algorithms. However, they still suffer from a fundamental limitation. Many rule-based classifiers used various greedy techniques to prune the redundant rules that lead to missing some important rules. Another challenge that must be considered is related to the enormous set of mined rules that result in high processing overhead. The result of these approaches is that the final selected rules may not be the global best rules. These algorithms are not successful at exploiting search space effectively in order to select the best subset of candidate rules. We merged the Apriori algorithm, harmony search, and classification based association rules (CBA) algorithm for building a rule-based classifier. We applied a modified version of the Apriori algorithm with multiple minimum support for extracting useful rules for each class in the dataset. Instead of using a large number of candidate rules, binary harmony search was utilized for selecting the best subset of rules that appropriate for building a classification model. We applied the proposed method on seventeen benchmark dataset and compared its result with traditional association rule classification algorithms. The statistical results show that our proposed method outperformed other rule-based approaches.