Item Features Research Articles

Enhancing Ecommerce Recommenders with Dual Transformer Models

The growing complexity and scale of ecommerce platforms have created a need for more advanced and precise recommendation systems. Traditional recommendation approaches, such as collaborative filtering and content-based models, while effective, often struggle to fully capture the evolving preferences and behaviours of users in dynamic online marketplaces. To address these limitations, this paper explores the application of dual transformer models in enhancing ecommerce recommender systems. Transformer models, originally developed for natural language processing, offer powerful attention mechanisms that allow for the processing of sequential and contextual data, providing a more nuanced understanding of user behaviour. This study introduces dual transformer models, which simultaneously process both user interaction history and item features to improve recommendation accuracy. By capturing the intricate relationships between users and items from both dimensions, dual transformer models offer a bidirectional approach that can adapt to changing user preferences in real time. This enables ecommerce platforms to deliver more relevant, personalized, and context-aware product recommendations. The use of dual transformers in recommendation systems presents several benefits, including improved scalability, enhanced personalization, and increased user engagement. By addressing the shortcomings of traditional recommender systems, dual transformer models have the potential to revolutionize the ecommerce industry, leading to better user satisfaction and higher conversion rates. This paper highlights the potential of these models in improving the overall effectiveness of ecommerce recommendation engines, making them a valuable tool for modern online shopping platforms.

Neural Embedding Singular Value Decomposition for Collaborative Filtering.

Singular value decomposition (SVD) is one of the most effective algorithms in recommender systems (RSs). Due to the iterative nature of SVD algorithms, one big challenge is initialization that has a major impact on the convergence and performance of RSs. Unfortunately, existing SVD algorithms in the literature typically initialize the user and item features in a random manner; thus, data information is not fully utilized. This work addresses the challenge of developing an efficient initialization method for SVD algorithms. We propose a general neural embedding initialization framework, where a low-complexity probabilistic autoencoder neural network initializes the features of user and item. This framework supports explicit and implicit feedback data sets. The design details of our proposed framework are elaborated and discussed. Experimental results show that RSs based on our proposed initialization framework outperform the state-of-the-art methods in rating prediction. Moreover, regarding item ranking, our proposed framework shows an improvement of at least 2.20% ~5.74% than existing SVD algorithms and other matrix factorization methods in the literature.

Neural Time-Aware Sequential Recommendation by Jointly Modeling Preference Dynamics and Explicit Feature Couplings.

In recommendation, both stationary and dynamic user preferences on items are embedded in the interactions between users and items (e.g., rating or clicking) within their contexts. Sequential recommender systems (SRSs) need to jointly involve such context-aware user-item interactions in terms of the couplings between the user and item features and sequential user actions on items over time. However, such joint modeling is non-trivial and significantly challenges the existing work on preference modeling, which either only models user-item interactions by latent factorization models but ignores user preference dynamics or only captures sequential user action patterns without involving user/item features and context factors and their coupling and influence on user actions. We propose a neural time-aware recommendation network (TARN) with a temporal context to jointly model 1) stationary user preferences by a feature interaction network and 2) user preference dynamics by a tailored convolutional network. The feature interaction network factorizes the pairwise couplings between non-zero features of users, items, and temporal context by the inner product of their feature embeddings while alleviating data sparsity issues. In the convolutional network, we introduce a convolutional layer with multiple filter widths to capture multi-fold sequential patterns, where an attentive average pooling (AAP) obtains significant and large-span feature combinations. To learn the preference dynamics, a novel temporal action embedding represents user actions by incorporating the embeddings of items and temporal context as the inputs of the convolutional network. The experiments on typical public data sets demonstrate that TARN outperforms state-of-the-art methods and show the necessity and contribution of involving time-aware preference dynamics and explicit user/item feature couplings in modeling and interpreting evolving user preferences.

Deep Recommendation With Adversarial Training

Recommender systems provide an effective solution to the information overload and have become a research hotspot in both industry and academia. Some existing work on implicit feedback has made great progress. However, there are still the following shortcomings: 1) existing pairwise ranking methods mostly sample unobserved items uniformly to obtain negative samples, which brings a biased solution and insufficient convergence; 2) the recommendation result comes from a complicated process, which makes it less interpretable. Therefore, we put forward a Deep Recommendation with Adversarial Training (DRAT) model, which provides users with personalized recommendations by utilizing an encoder-decoder structure and adversarial training. It consists of two components: one is feature learning module in which the encoder captures the text features of items from user-generated reviews, and the decoder reconstructs the text with the captured features; the other is rating prediction module which utilizes adversarial training to generate personalized negative samples tackling the drawbacks of uniform negative sampling. Extensive experiments on five real-world datasets show that our proposed model significantly outperforms the baseline methods.

Handling Dynamic User Preferences Using Integrated Point and Distribution Estimations in Collaborative Filtering

Change in users’ preferences over time is a challenging issue in collaborative filtering models. Recent temporal recommender systems (RSs) use latent temporal features or probabilistic latent transition models. The models designed using latent temporal features remain confined to point estimation only, i.e., considering single predicted values (e.g., rating) and completely ignore the likelihood of other possible items being generated due to change in users’ preferences. The latent transitive models represent users’ preference over different states using distribution estimation but ignore user–item features. This article proposes an integrated approach for designing collaborative filtering models using both point and distribution estimation techniques together to address this issue. The matrix factorization-based point estimation strategy is applied to integrate user–item correlations and feature information, whereas the hidden Markov model is used for users’ preference distribution over time. Three benchmark datasets were used to evaluate and compare the proposed model’s performance with related state-of-the-art temporal RS models. The proposed model makes the system more robust in handling dynamic users’ preferences.

Developing progress monitoring measures: Parallel test construction from the item-up

Progress monitoring is a process of collecting ongoing samples of student work and tracking performance of individual students over time. Progress monitoring involves administering parallel sets of items to the same student on a regular basis (at least monthly) that are sensitive to changes in the student’s understanding based on instruction. The sets of items administered over time should be parallel in difficulty so that differences in performance can be attributed to differences in the student’s understanding as opposed to variability in item difficulty across sets. In this manuscript, we describe an approach to designing items that controls item-level variability by constraining the item features that may elicit different cognitive processing. This approach adapts the principles of Automated Item Generation (AIG) and includes carefully designing test specifications, isolating specific components of the content that will be assessed, creating item models to serve as templates, duplicating the templates to create parallel item clones, and verifying that the duplicated item clones align with the original item model. An example from an operational progress monitoring system for mathematics in Kindergarten through Grade 6 is used to illustrate the process. We also propose future studies to empirically evaluate the assertion of parallel form difficulty.

Effectiveness of Semantics Feature Analysis on EFL Primary Learners’ Word Awareness

The correlation between the thematic and lexical units is a fundamental stage in language development. Understanding vocabulary from any text and category is essential in school subjects and daily life. Therefore, the current study aims at examining and analyzing the effect of semantic feature analysis (henceforth, SFA) strategy on the word awareness of EFL primary learners. The present study participants are twenty-three learners with language delay (LD), who are learning English as a second language at Da Vinci International School and Smart Kurd Private School, Duhok city, Kurdistan Region of Iraq. Furthermore, to achieve the research aims, a qualitative methodology was used, whereby students were observed, and data was collected through two main tasks, a matrix chart with and without graphical representation. The results showed that students faced fewer difficulties when identifying the physical features of lexical items in both matrixes. By contrast, the participants tended to find location features as the most difficult category. The study has also revealed that SFA with graphic representation is greatly effective in terms of vocabulary learning and retention. Moreover, this study is valuable as it investigated the effectiveness of SFA on primary learners’ word awareness. It contributes largely to children's language development in terms of lexicon enrichment, comprehension, and content retention.

An Improved Web Service Recommendation Method based on Decomposition Machine Model

With the increasing number of Web services with similar functions on the Internet, traditional collaborative filtering service recommendation methods may encounter problems such as data sparseness, cold start, and poor scalability. To solve the above problems, this project proposes a new Web service recommendation method based on the decomposition machine model. The method decomposes the user trust relationship matrix and the product rating matrix while adding the geographic location information of the service, and transforms the correlation matrix of the calculated user feature vector and item feature vector into the same latent factor space by means of a decomposition machine. Optimize training model parameters to provide users with accurate prediction scores. The ultimate purpose of QoS prediction is to recommend high-quality services to users, improve the efficiency of users' discovery and selection of high-quality services, and ultimately promote the utilization of network Web services and promote service providers to release higher-quality services. Scientific significance, but also has better application value.

How Long-Term English Learners Perform on an English Language Proficiency Assessment During Grades 2 through 5: An Examination of Assessment Tasks and Features

Using two states’ longitudinal data, this study tracked long-term English learner (LTEL) students’ performance on an annual English language proficiency (ELP) assessment (the ELPA21) from their pre-LTEL period. We followed English learner (EL) students who were initially classified as EL students in Kindergarten or in Grade 1 until they reached Grade 5. We defined LTEL students as EL students who have not achieved English language proficiency after five years of their initial EL classification. We examined LTEL students’ overall-, domain-, task-, and item-level performance in the ELPA21 and compared how their performances differed from non-LTEL students. We also identified features of items in the Grades 2 through 5 ELPA21 that were particularly challenging for LTEL students. LTEL students showed significantly lower proficiency and slower development in reading and writing than in listening and speaking domains in Grades 2 through 5, suggesting the need for strengthening elementary EL services and targeting instruction for EL students at risk of becoming LTEL students. In addition, findings suggest variability of task types and item features across grade levels that presented more challenges for LTEL students. The findings have implications for EL policies, programs, and instruction, especially for elementary EL students at risk of becoming LTEL students.

Constraints of attention, stimulus modality, and feature similarity in working memory.

Two of the most important concepts in working memory that differentiate many theories are the role of attention and similarity between items. Investigators have debated whether there is a central, general resource of attention, and whether interference between items depends mostly on their modality and type of code (i.e., verbal/acoustic versus visual/spatial coding) or upon multiple dimensions of feature similarity even within a modality. Here, we examine results from three experiments in which the features of items to be remembered differed for visual objects in color or orientation, or for acoustic objects in noise duration or tone pitch. There were one or two of these sets on a trial and, when there were two sets, the similarity between their features varied: there were sets in different modalities, sets with different feature types within a modality, or sets of the same feature type. One-set trials consistently produced superior performance. For two-set trials, dissimilarity of the sets mattered only when both sets had to be attended, compared with attention to only one set. Feature differences within a modality mattered at least as much as between-modality differences. The findings conflict with what would be expected if modality were the sole organizing principle and support a working memory model in which a capacity-limited attention is constrained by the feature similarity of task-relevant items.

A fuzzy content-based group recommender system with dynamic selection of the aggregation functions

Recommender systems are currently software tools that are focused on providing users with the best choices in an overloaded search space of possible options. Hence, group recommender systems have recently become an important trend in recommendation, because they aim at recommending a special type of items so-called social items, that tend to be consumed in groups such as TV programs, travel packages, etc. Among the different types of algorithms applied for group recommender systems, this paper is focused on content-based group recommender systems, as a novel group recommendation paradigm that exploits item features in the recommendation generation process. Specifically, our goal is to introduce a new content-based group recommendation approach, based on the recommendation aggregation paradigm whose main novelty is the development of a dynamic selection process of the aggregation scheme. Such an approach is centered on the identification of group's characteristics that are matching with the most appropriate function to use in the individual recommendation aggregation step. To perform such a matching, it is proposed a fuzzy decision tree induction process. The experimental evaluation shows that this scheme improves the recommendation performance of previous content-based group recommendation approaches, as well as it serves a starting point for further research based on this dynamic selection paradigm.

Identification of Casting Product Surface Quality Using Alex net and Le-net CNN Models

Casting is a manufacturing process in which a fluid product is normally poured into a mold, which contains a hollow cavity of the preferred form, and afterwards allowed to solidify. A Casting defect is an undesirable abnormality in a metal casting procedure. There are lots of types of issues in casting like blow openings, pinholes, burr, shrinkage issues, product flaws. Defects are an undesirable thing in the casting Industry. In this job, we extracted different casting items features and then applied convolutional neural network-based models for the discovery of the casted item is great or otherwise. So, it observed that neural networks can record the colours as well as textures of casting particularly to respective, which looks like human decision-making. This design is to deploy the Django internet framework. We try out different surfaces as input to convolutional neural networks for the efficient classification of the surface defect.

Knowledge Graph Recommendation Model Based on Feature Space Fusion

The existing recommendation model based on a knowledge graph simply integrates the behavior features in a user–item bipartite graph and the content features in a knowledge graph. However, the difference between the two feature spaces is ignored. To solve this problem, this paper presents a new recommendation model named the knowledge graph recommendation model based on feature space fusion (KGRFSF). Specifically, in the behavioral feature space, the behavioral features of users and items are constructed by extracting the behavioral feature from the user–item bipartite graph. In the content feature space, the content features related to users and items are extracted through the attention mechanism on the knowledge graph, and then the content feature vectors of users and items are constructed. Finally, through the feature space fusion model, the behavior features and content features are projected into the same preference feature space, and then the fusion of the two feature spaces is completed to construct the complete vector representations of users and items and calculate the vector similarity to predict the score of the user to the item. This paper applies the presented model to public datasets in the fields of music and film. It can be found through the experimental results that KGRFSF can effectively improve the recommendation accuracy compared with the existing models.

Hybrid recommender system with core users selection

Recommender system plays an increasingly important role in identifying the individual’s preference and accordingly makes a personalized recommendation. Matrix factorization is currently the most popular model-based collaborative filtering (CF) method that achieves high recommendation accuracy. However, similarity computation hinders the development of CF-based recommendation systems. Preference obtained only depends on the explicit rating without considering the implicit content feature, which is the root cause of preference bias. In this paper, the content feature of items described by fuzzy sets is integrated into the similarity computation, which helps to improve the accuracy of user preference modeling. The importance of a user is then defined according to preferences, which serves as a baseline standards of the core users selection. Furthermore, core users based matrix factorization model (CU-FHR) is established, then genetic algorithm is used to predict the missing rating on items. Finally, MovieLens is used to test the performance of our proposed method. Experiments show CU-FHR achieves better accuracy in prediction compared with the other recommendation methods.

Deep transfer learning with multimodal embedding to tackle cold-start and sparsity issues in recommendation system.

Recommender systems (RSs) have become increasingly vital in the modern information era and connected economy. They play a key role in business operations by generating personalized suggestions and minimizing information overload. However, the performance of traditional RSs is limited by data sparseness and cold-start issues. Though deep learning-based recommender systems (DLRSs) are very popular, they underperform when considering rating matrices with sparse entries. Despite their performance improvements, DLRSs also suffer from data sparsity, cold start, serendipity, and generalizability issues. We propose a multistage model that uses multimodal data embedding and deep transfer learning for effective and personalized product recommendations, and is designed to overcome data sparsity and cold-start issues. The proposed model includes two phases. In the first—offline—phase, a deep learning technique is implemented to learn hidden features from a large image dataset (targeting new item cold start), and a multimodal data embedding is used to produce dense user feature and item feature vectors (targeting user cold start). This phase produces three different similarity matrices that are used as inputs for the second—online—phase to generate a list of top-n relevant items for a target user. We analyzed the accuracy and effectiveness of the proposed model against the existing baseline RSs using a Brazilian E-commerce dataset. The results show that our model scored 0.5882 for MAE and 0.4011 for RMSE which is lower than baseline RSs which indicates that the model achieved an improved accuracy and was able to minimize the typical cold start and data sparseness issues during the recommendation process.

Conversational recommendation: Theoretical model and complexity analysis

Recommender systems help users find items of interest in situations of information overload in a personalized way, using needs and preferences of individual users. In conversational recommendation approaches, the system acquires needs and preferences in an interactive, multi-turn dialog. This is usually driven by incrementally asking users about their preferences about item features or individual items. A central research goal in this context is efficiency, evaluated concerning the number of required interactions until a satisfying item is found. Today, research on dialog efficiency is almost entirely empirical, aiming to demonstrate, for example, that one strategy for selecting questions to ask the user is better than another one in a given application. This work complements empirical research with a theoretical, domain-independent model of conversational recommendation. This model, designed to cover a range of application scenarios, allows us to investigate the efficiency of conversational approaches in a formal way, particularly concerning the computational complexity of devising optimal interaction strategies. An experimental evaluation empirically confirms our findings.

Unexpected interest recommender system with graph neural network

AbstractTraditional recommender systems often face the filter bubble problem when they focus on recommending familiar items to users. The over-specialized recommended contents will make users bored. To solve this problem, researchers have proposed models that focus on unexpectedness, but these models all suffer from incomplete learning of features. To address this problem, we propose an unexpected interest recommender system with graph neural network (UIRS-GNN). First, we preprocess the input data with a graph convolutional network. It enriches user and item feature vectors by aggregating neighborhood information. Second, we transform the GRU and propose the attention-based long short-term gated recurrent unit network to learn user preferences hidden in historical behavior sequences. Then, we input the preprocessed feature vectors of users and items into the unexpected interest model, and solve the problem of insufficient feature information learning by aggregating neighborhood information. Furthermore, our model also alleviates data sparsity due to our deep learning feature information. Finally, empirical evaluations with several competitive baseline models on three real-world datasets reveal the superior performance of UIRS-GNN.

Does a rabbit have feathers or fur? Development of a 42-item semantic memory test (SMT-42)

ABSTRACT Objective We present the preliminary study of the 42-item Semantic Memory Test (SMT-42), a test developed to distinguish semantic variant primary progressive aphasia (svPPA) from the other variants: logopenic (lPPA) and nonfluent/agrammatic (naPPA). The test requires the patient to retrieve the conceptual features of items belonging to different lexical categories. Methods In the first study, we administered the French version of the SMT-42 to a population of healthy subjects and to patients with svPPA matched to a subgroup of the healthy subjects. In the second study, we administered the SMT-42 to four groups of patients (with svPPA, lPPA, naPPA and Alzheimer’s disease [AD], respectively) to study its capacity to differentiate patients suffering from svPPA from the other patients. Results In the first study, 109 healthy subjects were included, 15 of whom were paired with 15 subjects presenting with svPPA. In the second study, designed to compare groups presenting a primary progressive aphasia variant and AD, 12 subjects with svPPA, 6 with naPPA and 9 with lPPA were included, along with 21 subjects with AD. The subjects presenting a semantic deficit were clearly distinguished from the others by their results on the SMT-42 (svPPA: mean = 30.0 (5.9); lPPA: mean = 37.8 (3.3), d = 1.5, p = 0.002; naPPA: mean = 39.8 (1.9), d = 1.89, p = 0.001; AD: mean = 38.5 (2.4), d = 1.63, p < 0.001); (svPPA: median = 31; lPPA: median = 38, U = 9, p = 0.002; naPPA: median = 40.5, U = 1.5, p = 0.001; AD: median = 39, U = 13.5, p < 0.001). Conclusion The SMT-42 is simple, rapidly administered (3 minutes on average), easily scored and has good sensitivity, and it appears to be an effective tool for semantic screening in routine clinical practice.

Comparative Convolutional Dynamic Multi-Attention Recommendation Model.

Recently, an attention mechanism has been used to help recommender systems grasp user interests more accurately. It focuses on their pivotal interests from a psychology perspective. However, most current studies based on it only focus on part of user interests; they have not mined user preferences thoroughly. To address the above problem, we propose a novel recommendation model: comparative convolutional dynamic multi-attention (CCDMA). This model provides a more accurate approach to represent user and item features and uses multi-attention-based convolutional neural networks to extract user and item latent feature vectors dynamically. The multi-attention mechanism considers both self-attention and cross-attention. Self-attention refers to the internal attention within users and items; cross-attention is the mutual attention between users and items. Moreover, we propose an optimized comparative learning framework that can mine the ternary relationships between one user and a pair of items, focusing on their relative relationship and the internal link between a pair of items. Extensive experiments on several real-world data sets show that the CCDMA model significantly outperforms state-of-the-art baselines in terms of different evaluation metrics.

Children expect others to prefer handmade items.

Although children frequently engage in creative activities (in which they make foods and objects by hand), the development and scope of children's thinking about handmade items is largely unexplored. In the present studies, we examined whether 4- to 12-year-old children at a local children's museum (54% girls, 46% boys; 51% White, 11% Asian/Asian American, 10% more than 1 group, 4% Latinx, 3% Black/African American, 18% did not report race/ethnicity) would expect other people to prefer handmade over factory-made items, including foods and nonfoods. In Experiments 1 (n = 124) and 2 (n = 122), participants expected a child character to prefer items the character made themselves and items made by the character's parent or a local person. However, this expectation did not persist at all costs: When considering imperfect handmade items in Experiment 3 (n = 122), children demonstrated a handmade preference when considering nonfoods made by a parent but demonstrated a factory-made preference when considering foods made by a parent. Children's explanations were associated with their choices: When children's explanations referred to emotions or relationships, they were more likely to select handmade items. When children referred to item features, they were more likely to select factory-made items. Across studies, we observed persistent age and gender effects: Children's handmade preference increased with child age and girls demonstrated a more robust handmade preference than boys. These findings highlight children's developing and nuanced reasoning about object value. At an early age, children consider who made an object as a contributor to its value. (PsycInfo Database Record (c) 2022 APA, all rights reserved).