Similar Items Research Articles

Enhancing performance factor analysis through skill profile and item similarity integration via an attention mechanism of artificial intelligence

IntroductionFrequent formative assessment is essential for accurately evaluating student learning, enhancing engagement, and providing personalized feedback. In STEM education, understanding the relationship between skills that students have internalized (mastered) and those they are developing (emergent) is crucial. Traditional models, including item response and cognitive diagnosis models, primarily focus on emergent skills, often overlooking internalized skills. Moreover, new tools like large language models lack a complete approach for tracking knowledge and capturing complex skill relationships.MethodsThis study incorporates artificial intelligence, specifically attention mechanisms, into educational assessment to evaluate both emergent and internalized skills. We propose a modified version of Performance Factor Analysis (PFA), which assesses student abilities by analyzing past responses and comparing them with peer performance on the same items, using parameters from a sigmoid function. This model leverages attention mechanisms to capture item order-based similarity and decay principles, providing a nuanced view of student skill profiles.ResultsThe Modified Performance Factor Analysis model significantly improved discriminative power, accuracy, precision, recall, and F1 scores across various skill areas compared to traditional PFA models.DiscussionThese results indicate that the Modified Performance Factor Analysis model allows for a more accurate and comprehensive evaluation of student performance, effectively identifying both emergent and internalized skills. By integrating AI into assessment, educators gain deeper insights, enabling them to refine teaching strategies and better support students' mastery of both types of skills.

Parental Stress in Autistic Children with Poor Oral Hygiene: A Pilot Study to Develop and Validate a Measurement Scale

Background/Objectives: Research indicates that children with autism spectrum disorder (ASD) exhibit a deficiency in skills and initiative when it comes to adhering to daily oral hygiene routines. This, in turn, increases the likelihood of oral pathologies, thereby placing a significant emotional strain on their parents. In addition to the typical stress they already experience, parents of children with ASD are also burdened with pediatric oral health issues. However, a review of the literature reveals a lack of studies measuring stress in parents of children whose oral health affects their autistic condition. This research aimed to design and to validate a stress scale for parents of autistic children with poor oral hygiene. Methods: The study used an exploratory, non-experimental design to validate and determine the reliability of the scale. Scale reliability was assessed through a cross-sectional pilot test to evaluate internal scale consistency, with a focus on item similarity. We used Aiken’s V to estimate the validity of the scale, and Cronbach’s α for calculating scale reliability. Calculations, estimations, and statistical analysis were conducted using SPSS. Results: Expert validation, a pilot test, and a cross-sectional, non-experimental design established the reliability of the scale. Conclusions: We conclude that the 20-item scale exhibits validity (0.95) and reliability (0.965), ensuring its applicability in future research.

IBCFaiCDR: Auxiliary data-driven item-based collaborative filtering in cross-domain RSs to address user cold start problem

The adoption of Recommender Systems (RSs) in various services is imperative in the present era. One of the inherent challenge faced by RSs is the user cold start problem (UCS) and it occurs when there is a lack of enough data available for a new user, which restricts the capacity to provide accurate recommendations. Researchers have conducted investigations into the utilization of auxiliary data like social network data, review text, demographic information etc., to augment the RSs ability to learn and understand the user's preferences, with the aim of solving this issue. The provided information is inadequate due to the restricted level of user engagement within the specified domain. The researchers utilize a Cross-Domain RS (CDR) that incorporates information from multiple domains to provide personalized suggestions to users. Item Based Collaborative Filtering (IBCF), which utilizes item-item similarity, is commonly used when dealing with sparse data. The suggested collaborative filtering technique employs IBCF which uses the item similarity obtained with item features. These features are generated by the proposed approach IBCFaiCDR's model, trained on source domain data such as product images and reviews. The efficacy of the proposed approach is assessed using a large dataset from Amazon, consisting of user interactions across many categories. The results on target domain, evaluated using MAE and RMSE, indicate that the inclusion of auxiliary information (image + text) enhances the RS performance for new users.

Pattern separation during encoding and Subsequent Memory Effect

Memory retrieval has been extensively studied in relation to the encoding processes that precede access to stored information. Event related potentials (ERP) research has compared brain potentials elicited during the study phase of successful and unsuccessful retrieval, finding greater activation for the subsequent retrieval information. In this work we were interested in exploring the neural markers associated to subsequent recognition when similar memories are subsequently encoded. We used a Subsequent Memory paradigm in which we manipulated the number of similar items within a category (2 or 6) that participants encoded. Manipulating the number of similar encoded items within a category allowed us to test whether encoding markers of subsequent recognition depend solely on memory trace strength or, on the contrary, successful recognition is influenced by subsequently presented similar memories, and consequently may not be reflected in higher activation in such cases. After a 20-minute period, participants performed a recognition task providing one of a three-option judgement: “old”, “similar” and “new”, which allowed us to test if the amplitude of ERP waveforms varied based on the similarity judgement of the unrecognized encoded item. We did not observe a significant parietal subsequent memory effect, however, old hits and similar false alarms were both significantly different from similar correct rejections and old false alarms in ERP retrieval. These findings suggest that differences in brain responses between conditions are specifically related to the retrieval process and not the encoding process, indicating potential differential effects on memory during retrieval. Moreover, it is also possible that differences in brain responses develop or change over the rest time between phases, influencing how these conditions manifest across different stages of information processing.

Multi-interest sequential recommendation with contrastive learning and temporal analysis

Sequential recommendation systems aim to forecast the subsequent item of interest to users by analyzing their historical behaviors. While existing approaches, which employ attention mechanisms, have significantly advanced by capturing users’ multiple interests, they encounter two primary challenges. Firstly, they often fail to effectively capture the transient shifts in users’ interests across a sequence of items and neglect the interdependencies among these items, leading to a misalignment between the identified and actual interests. Secondly, conventional multi-interest models struggle to ensure that the identified interests are distinct, which results in overly similar interests that may not adequately satisfy user requirements. To address these issues, we propose a novel multi-interest recommendation method, which models the temporal features and user’s preference features from the user level. In order to capture short-term variations in interest, we introduce a time period module to encode the behavioral intervals between items and capture the periodicity of users clicking on similar items by extracting temporal information. In addition, we integrate similar types of items into the interest subgraph through preference feature extraction to capture users’ short-term changes in relevance term interests, and incorporate contrastive learning to enhance the differences between the captured interests. Extensive experiments conducted on two datasets Amazon Books and Taobao show that the model outperforms current state-of-the-art methods.

MCAP: Low-Pass GNNs with Matrix Completion for Academic Recommendations

Graph Neural Networks (GNNs) are commonly used and have shown promising performance in recommendation systems. A major branch, Heterogeneous GNNs, models heterogeneous information by leveraging side information for academic paper recommendations. These networks use message passing and high-order propagation to learn representations for users and items. However, existing recommendation methods perform high-order propagation, leading to suboptimal representation learning. To address this issue, this paper proposes a framework called MCAP, which uses relation-aware GNNs and executes low-pass propagation with matrix completion to enhance academic paper recommendations. The framework uses an attention mechanism to learn top- U relationships by constructing a user-user relation graph based on common authors and venues from interacted items. To efficiently and effectively capture semantic-aware similar items, MCAP builds an item-item relation graph by fusing side information of papers using text embedding models (e.g., Mistral) and large language models (e.g., GPT-3.5-Turbo, GLM-4). Finally, the relation-aware user-user and item-item graphs are incorporated into existing GNN-based models to generate representations of users and papers to enhance academic paper recommendations. The effectiveness of the MCAP is validated using four academic datasets, AMiner-PC, AMiner-WeChat, CiteULike, and DBLP, with user-item interactions and side information of papers. Comprehensive experiments show that the MCAP outperforms state-of-the-art models in terms of Recall@5, NDCG@5, and HR@5 with 69.2%, 70.5%, and 77.6% on the AMiner-WeChat dataset. The code for MCAP is available at https://github.com/THUDM/MCAP .

Towards similar alignment and unique uniformity in collaborative filtering

Representation learning, with its desired properties, including Alignment and Uniformity, has recently emerged as an effective approach in collaborative filtering (CF) for recommender systems. Alignment measures the distance between positive pairs, and uniformity describes the distribution of all samples on the unit hypersphere. Despite the effectiveness of existing research, the challenges of insufficient alignment and uniformity bias persist in studies on the desired properties of recommender systems: (1) Sparse interaction information leads to an insufficient alignment representation. (2) Calculating the uniformity loss based on duplicate samples introduces bias, which affects the overall representation uniformity.Building on aforementioned challenges, we propose a Similar Alignment and Unique Uniformity model called SUAU. SUAU mitigates insufficient alignment by introducing additional item-similar item pairs for model training and employing a uniqueness strategy to prevent uniformity bias. More specifically, we propose a similar items generation method that identifies the most similar items based on the interaction behavior of the original items’ related users to form item-similar item pairs. The experimental results obtained on three highly sparse datasets demonstrate that the propose model achieves superior alignment and uniformity compared with state-of-the-art models, notably surpassing the best CF methods in terms of recommendation performance. We have open-sourced the source code on a publicly accessible website: https://github.com/ZzYUuuu/SUAU.

Hybrid attribute-based recommender system for personalized e-learning with emphasis on cold start problem

This article introduces a recommendation system that merges a knowledge-based (attribute-based) approach with collaborative filtering, specifically addressing the challenges of the pure-cold start scenario in personalized e-learning. The system generates learning recommendations by assessing item similarities, utilizing the Rogers-Tanimoto similarity measure for materials and users, and Jaccard's similarity for user comparisons. Unlike traditional collaborative methods relying on prior ratings, this approach depends on attributes. Additionally, user and learning material profiling structures were created to serve as fundamental inputs for the recommendation algorithm. These profiles represent student and material knowledge in a two-dimensional space to facilitate matching. Our processes incorporate user learning styles, preferences, and prior knowledge as metrics for achieving the desired level of personalization. The system produces a list of top recommendations based on predicted ratings. To validate its efficacy, a website resembling our learning platform was developed and tested by users. The primary results demonstrate the system's ability to identify similar users even in a pure cold start condition without existing ratings. Consequently, the system proves its capability in recommending suitable materials, modeling students, and identifying similar user groups. The evaluation results of the proposed system showed a good level of satisfaction by the testimonials, quantified by a score of 82% for the recommended materials (16% higher than exiting cold-start systems), and an average score of 90% in terms of satisfaction about the generated student profiles. As they proved the capability of the framework in recommending suitable materials, and its capability in modeling students, finding similar groups of users.

RGMeta: Enhancing Cold-Start Recommendations with a Residual Graph Meta-Embedding Model

Traditional recommendation models grapple with challenges such as the scarcity of similar user or item references and data sparsity, rendering the cold-start problem particularly formidable. Meta-learning has emerged as a promising avenue to address these issues, particularly in solving the item cold-start problem by generating meta-embeddings for new items as their initial ID embeddings. This approach has shown notable success in enhancing the accuracy of click-through rate predictions. However, prevalent meta-embedding models often focus solely on the attribute features of the item, neglecting crucial user information associated with it during the generation of initial ID embeddings for new items. This oversight hinders the exploitation of valuable user-related information to enhance the quality and accuracy of the initial ID embedding. To tackle this limitation, we introduce the residual graph meta-embedding model (RGMeta). RGMeta adopts a comprehensive approach by considering both the attribute features and target users of both old and new items. Through the integration of residual connections, the model effectively combines the representation information of the old neighbor items with the intrinsic features of the new item, resulting in an improved initial ID embedding generation. Experimental results demonstrate that RGMeta significantly enhances the performance of the cold-start phase, showcasing its effectiveness in overcoming challenges associated with sparse data and limited reference points. Our proposed model presents a promising step forward in leveraging both item attributes and user-related information to address cold-start problems in recommendation systems.

The “Pink Tax” and Gender Price Disparity in Personal Care

The “Pink Tax” refers to charging higher prices for products and services marketed primarily to women compared to similar or identical items targeted at men, despite comparable production costs. This research uses Unilever as a case study to focus on the impact of the Pink Tax on women’s purchasing power, particularly within the personal care industry. The study provides a comprehensive overview of the Pink Tax’s evolution, key research findings, public responses, and regulatory measures to address this issue. By synthesising existing research and case studies, this paper highlights the economic burden imposed on women and the reinforcement of gender stereotypes through market practices. The study underscores the need for equitable pricing mechanisms and informed consumer advocacy to rectify the economic disparities exacerbated by the Pink Tax.

Disentangling Selection into Mode from Mode Effects.

We investigate the impact of data collection mode on responses to variables in NSHAP Round 4 and discuss how potential mode differences should (and should not) be addressed in substantive analyses. Among the set of respondents who were eligible to be contacted remotely in Round 4, we randomly selected 398 to be contacted instead for an in-person interview. We compare response rates and the distributions of selected key outcomes among those 398 respondents to those among the control group who were initially approached remotely. As a contrast, we compare all R4 respondents according to the mode in which they completed the interview, including those not part of the randomized experiment. Among those included in the experiment, there was no evidence of systematic differences in responses to physical and mental health questions between remote and in-person modes, nor in responses to number recall measures. In-person respondents scored moderately lower on cognitive function measures requiring careful attention to a figure and/or task, though this difference became less with each similar item. Remote respondents named fewer social network members. Comparing all respondents according to their final mode yielded substantially different results in all cases. Mode did not appear to affect reports of physical and mental health based on a randomized comparison, though it did moderately affect other items in predictable ways. Naïve estimates of mode effects based on comparing all respondents according to mode yielded misleading results, and should not be used to adjust for mode differences in analyses.

Subsequent Memory Effects in Cortical Pattern Similarity Differ by Semantic Class.

Although living and nonliving stimuli are known to rely on distinct brain regions during perception, it is largely unknown if their episodic memory encoding mechanisms differ as well. To investigate this issue, we asked participants to encode object pictures (e.g., a picture of a tiger) and to retrieve them later in response to their names (e.g., word "tiger"). For each of four semantic classes (living-animate, living-inanimate, nonliving-large, and nonliving-small), we examined differences in the similarity in activation patterns (neural pattern similarity [NPS]) for subsequently remembered versus forgotten items. Higher NPS for remembered items suggests an advantage of within-class item similarity, whereas lower NPS for remembered items indicates an advantage for item distinctiveness. We expect NPS within class-specific regions to be higher for remembered than for forgotten items. For example, the parahippocampal cortex has a well-known role in scene processing [Aminoff, E. M., Kveraga, K., & Bar, M. The role of the parahippocampal cortex in cognition. Trends in Cognitive Sciences, 17, 379-390, 2013], and the anterior temporal and inferior frontal gyrus have well-known roles in object processing [Clarke, A., & Tyler, L. K. Object-specific semantic coding in human perirhinal cortex. Journal of Neuroscience, 34, 4766-4775, 2014]. As such, we expect to see higher NPS for remembered items in these regions pertaining to scenes and objects, respectively. Consistent with this hypothesis, in fusiform, parahippocampal, and retrosplenial regions, higher NPS predicted memory for subclasses of nonliving objects, whereas in the left inferior frontal and left retrosplenial regions, lower NPS predicted memory for subclasses of living objects. Taken together, the results support the idea that subsequent memory depends on a balance of similarity and distinctiveness and demonstrate that the neural mechanisms of episodic encoding differ across semantic categories.

Digitalization for enhancing reading habits: the improved hybrid book recommendation system with genre-oriented profiles

PurposeReading habit plays a pivotal role in individuals' personal and academic growth, making it essential to encourage among campus users. University libraries serve as valuable platforms to promote reading by providing access to a diverse range of books and resources. Recommending books through personalized systems not only helps campus users discover new materials but also enhances their engagement and satisfaction with the library’s offerings, contributing to a holistic learning experience.Design/methodology/approachThis study presents a web-based solution, the Web-Based Hybrid Intelligent Book Recommender System (W_HybridBook), as a solution that addresses challenges like cold start issues and limited scalability by factoring in user preferences and item similarities in generating book recommendations. The paper improves the traditional hybrid system using Genre-Oriented Profiles (GOPs) instead of original rating profiles of users when determining similarities between individuals. Consumption-based genre profiles (W_HybridBook-CBP) are created by assessing whether an item has received any ratings in the dataset, and vote-based genre profiles (W_HybridBook-VBP) are generated by considering the genre categories based on the magnitude of the user’s rating.FindingsThe comparative results indicated that users are quite satisfied with the recommendations generated by W\\_HybridBook-VBP profiling, with an average rating of 4.0633 and a precision value of 0.7988. W\\_HybridBook-VBP is also the fastest way with respect to the algorithm and recommendation run time.Originality/valueThe proposed W\\_HybridBook has been then enhanced by adopting two user profiling strategies to boost the similarity calculation process in the recommendation generation phase. This system provides ranking-based recommendations by mainly integrating well-known collaborative and content-based filtering strategies. A dataset has been collected by considering the preferences of both users and academics at Izmir Bakircay University, which is one of the universities with the highest number of books per student. More importantly, this dataset has been released and become publicly available for future research in the recommender system field.

Comparison between AWGC-cachexia and GLIM-malnutrition in patients with gastric cancer

BackgroundThe newly released Asian Working Group for Cachexia (AWGC) criteria share similar diagnostic items with the Global Leadership Initiative on Malnutrition (GLIM) criteria. This study aims to compare the AWGC cachexia and GLIM malnutrition in patients with gastric cancer and investigate whether one diagnosis continues to be a prognostic factor in individuals diagnosed with the other condition. MethodsData of patients who underwent radical gastrectomy for gastric adenocarcinoma were prospectively collected from 2013 to 2019. The AWGC and GLIM criteria were applied to diagnosis cachexia and malnutrition, respectively. Univariate and multivariate logistic and Cox regression were used to verify the effect of relevant factors on postoperative complications and overall survival. ResultsA total of 1420 patients were included, among whom 174 (12.3 %) were diagnosed with AWGC-cachexia alone, 85 (6.0 %) were diagnosed with GLIM-malnutrition alone, and 324 (22.8 %) had both AWGC-cachexia and GLIM-malnutrition. Both AWGC-cachexia and GLIM-malnutrition were independent risk factors for complications and overall survival. When they coexisted, the odds ratios (OR) and hazard ratios (HR) tended to be higher. In the AWGC-cachexia subset, GLIM-malnutrition remained an independent risk factor (HR = 1.544, 95 % CI = 1.098–2.171, P = 0.012) for overall survival after the adjustment of confounding factors. Similarly, in the GLIM-malnutrition subset, AWGC-cachexia remained an independent risk factor for overall survival (HR = 1.697, 95 % CI = 1.087–2.650, P = 0.020). Patients with both cachexia and malnutrition had the worst overall survival. ConclusionAWGC-cachexia and GLIM-malnutrition criteria were two non-redundancy tools in reflecting mortality risk in preoperative nutritional assessment.

Ordered spatial configuration protects representations of dissimilar items and reduces the similarity effect in visual working memory.

Although similarity could improve visual working memory (VWM) performance, it remains unclear how the spatial configuration of visual information influences the similarity effect in VWM. We explored this question by manipulating the orderliness of spatial configuration (ordered vs. scrambled) in the simultaneous (Experiment 1) and sequential (Experiment 2) change detection tasks. The results showed that similarity improved VWM performance when memory items were presented simultaneously and sequentially. For the simultaneous memory array containing similar and dissimilar items, the performance of the ordered spatial configuration was better than that of the scrambled spatial configuration when probing dissimilar items, while no such difference was found when probing similar items. Further, the similarity effect value in the scrambled spatial configuration was higher than that in the ordered spatial configuration. For the sequential memory array containing similar and dissimilar items, spatial configuration did not affect the similarity effect in VWM. Taken together, these findings suggest that spatial configuration could modulate the similarity effect when memory items are presented simultaneously, in which the ordered spatial configuration protects representations of dissimilar items and reduces the similarity effect in VWM. Our study provides additional evidence for the role of spatial configuration in the similarity effect in VWM, and supports the hierarchical model.

Collaborative filtering with representation learning in the frequency domain

In the context of recommender systems, collaborative filtering is the method of predicting the ratings of a set of items given by a set of users based on partial knowledge of the ratings. Commonly, items and users are represented via vectors, and to predict ratings, approaches such as vector inner-product (aka matrix factorization) or more advanced nonlinear functions are applied. In this paper, while we adopt the common vectorial representation, we consider a general model in which the ratings are smooth functions of the item representations. Smoothness ensures similar items with nearby vectors will also get similar ratings as we expect from a human rater. We represent user smooth scoring functions in a so-called frequency domain and learn their representations alongside item representations using 1) an iterative optimization approach that maps items and users alternatively, and 2) a feedforward neural network consisting of interpretable layers. We also address the challenge of the distribution shift from observed to unobserved ratings (aka missing-not-at-random) with insights from the frequency domain. We evaluate the predictive power of our method and its robustness in missed-not-at-random settings on four popular benchmarks. Despite its simplicity and interpretability, our method yields a remarkable performance compared to the state-of-the-art.1

The Impact of Investor Expectation on the Financial Decision-Making

Unlike traditional finance, behavioral finance challenges the traditional financial theory of investor rationality, emphasizing that investors are not absolutely rational in real life. This perspective provides new insights for understanding and predicting financial markets. This study examines core concepts in behavioral finance, such as loss aversion, market anomalies, momentum effects, reversal effects, endowment effects, and framing effects, and analyzes how they influence investors' financial decisions. These concepts highlight the significant role of human psychology and behavior in shaping financial decisions, thus aiding investors in making more rational choices in complex financial markets. Loss aversion, for instance, suggests that investors are more sensitive to losses than to gains, leading to suboptimal decision-making. Market anomalies, such as price bubbles and crashes, indicate that markets are not always efficient. Momentum effects suggest that past stock performance can predict future trends, while reversal effects point to the tendency of extreme stock performances to revert to the mean. The endowment effect reveals that individuals value their own possessions more highly than similar items they do not own. Lastly, framing effects demonstrate how the presentation of information can influence decision-making. Future research can delve deeper into how the framing effect can be harnessed to improve the design and marketing strategies of financial products, thereby enhancing the efficiency and effectiveness of investment decisions. This exploration could lead to more sophisticated financial products and better-informed investors, ultimately contributing to a more stable and efficient financial market.

Using natural language processing to facilitate the harmonisation of mental health questionnaires: a validation study using real-world data

BackgroundPooling data from different sources will advance mental health research by providing larger sample sizes and allowing cross-study comparisons; however, the heterogeneity in how variables are measured across studies poses a challenge to this process.MethodsThis study explored the potential of using natural language processing (NLP) to harmonise different mental health questionnaires by matching individual questions based on their semantic content. Using the Sentence-BERT model, we calculated the semantic similarity (cosine index) between 741 pairs of questions from five questionnaires. Drawing on data from a representative UK sample of adults (N = 2,058), we calculated a Spearman rank correlation for each of the same pairs of items, and then estimated the correlation between the cosine values and Spearman coefficients. We also used network analysis to explore the model’s ability to uncover structures within the data and metadata.ResultsWe found a moderate overall correlation (r = .48, p < .001) between the two indices. In a holdout sample, the cosine scores predicted the real-world correlations with a small degree of error (MAE = 0.05, MedAE = 0.04, RMSE = 0.064) suggesting the utility of NLP in identifying similar items for cross-study data pooling. Our NLP model could detect more complex patterns in our data, however it required manual rules to decide which edges to include in the network.ConclusionsThis research shows that it is possible to quantify the semantic similarity between pairs of questionnaire items from their meta-data, and these similarity indices correlate with how participants would answer the same two items. This highlights the potential of NLP to facilitate cross-study data pooling in mental health research. Nevertheless, researchers are cautioned to verify the psychometric equivalence of matched items.

Why leave items in the shopping cart? The impact of consumer filtering behavior

Online product information provides crucial cues for consumer shopping behavior; however, the impact of consumer-side information manipulation on non-purchase behavior (e.g., shopping cart abandonment) remains unclear. Filtering, a common manipulation strategy, was initially considered synonymous with searching, both of which contribute to consumer purchase behavior. However, recent research suggests significant differences, particularly in terms of cost and product matching. This study explores how such differences affect information manipulation and subsequently shape shopping cart abandonment, drawing on information foraging theory. Based on the premise that a forager's interaction with the environment is mediated by cognition, we find that consumer filtering behavior influences perceptions of serendipity (i.e., discovering unexpected items) and similarity (i.e., encountering similar items), thereby influencing shopping cart abandonment. Mediation analyses reveal that perceived serendipity and similarity are equally potent mediators. Marginal effects analyses demonstrate that a one standard deviation increase in consumer perceptions of serendipity (similarity) in filtering results increases the likelihood of shopping cart abandonment by 32.2% (22.1%). Theoretical contributions and practical insights for information behavior researchers and digital marketers are also provided.

Research on sales strategies for target customers of electric vehicles based on gradient boosting ensemble trees

This paper focuses on the automotive industry, especially the new energy vehicle sector, which is vital for the economy and needs careful market sales decisions. It analyzes data from 1,964 potential customers, including their satisfaction and personal information. The study uses machine learning to create a math model and give sales advice. The data is first cleaned without changing its original structure, and then normalized using the Z-Score method. For any outlier values in the data, the mean of similar items is used to fill in gaps. Specifically, for 500 missing entries about the number of children (B7), the paper groups customers by age, calculates the frequency of children in each group, and uses these frequencies to fill in missing values.Using the cleaned data, the XGBoost algorithm ranks the factors influencing customer decisions. For Brand 1, the top factors are economy (a3), work experience (B10), and safety performance (a4). For Brand 2, comfort (a2), disposable income (B15), and mortgage expenses (B16) are key. For Brand 3, car loan expenses (B17), battery technology (a1), and again mortgage expenses (B16) are most important.The paper also addresses imbalanced data by using an XGBoost binary classification model. This model predicts the likelihood of buying an electric vehicle using the Sigmoid function on tree node weights. It evaluates the model using Precision, Recall, and Area Under the Receiver Operating Characteristic Curve (AUC), which works well even with imbalanced data. The Precision and Recall rates for all three brands are above 90%, with AUC over 85% for Brands 1 and 2, and over 75% for Brand 3. This shows that the model is effective and can be used in real-world situations.