Implicit Data Research Articles

Enhanced Prompt Learning for Few-shot Text Classification Method

An enhanced prompt learning method (EPL4FTC) for few-shot text classification task is proposed. This algorithm first converts the text classification task into the form of prompt learning based on natural language inference. Thus, the implicit data enhancement is achieved based on the prior knowledge of pre-training language models and the algorithm is optimized by two losses with different granularities. Moreover, to capture the category information of specific downstream tasks, the triple loss is used for joint optimization. The masked-language model is incorporated as a regularizer to improve the generalization ability. We evaluated our method on four Chinese and three English text classification datasets. The experimental results show that the classification accuracy of the proposed EPL4FTC is significantly better than the other compared baselines.

Generating implicit object fragment datasets for machine learning

One of the primary challenges inherent in utilizing deep learning models is the scarcity and accessibility hurdles associated with acquiring datasets of sufficient size to facilitate effective training of these networks. This is particularly significant in object detection, shape completion, and fracture assembly. Instead of scanning a large number of real-world fragments, it is possible to generate massive datasets with synthetic pieces. However, realistic fragmentation is computationally intensive in the preparation (e.g., pre-factured models) and generation. Otherwise, simpler algorithms such as Voronoi diagrams provide faster processing speeds at the expense of compromising realism. In this context, it is required to balance computational efficiency and realism. This paper introduces a GPU-based framework for the massive generation of voxelized fragments derived from high-resolution 3D models, specifically prepared for their utilization as training sets for machine learning models. This rapid pipeline enables controlling how many pieces are produced, their dispersion and the appearance of subtle effects such as erosion. We have tested our pipeline with an archaeological dataset, producing more than 1M fragmented pieces from 1,052 Iberian vessels (). Although this work primarily intends to provide pieces as implicit data represented by voxels, triangle meshes and point clouds can also be inferred from the initial implicit representation. To underscore the unparalleled benefits of CPU and GPU acceleration in generating vast datasets, we compared against a realistic fragment generator that highlights the potential of our approach, both in terms of applicability and processing time. We also demonstrate the synergies between our pipeline and realistic simulators, which frequently cannot select the number and size of resulting pieces. To this end, a deep learning model was trained over realistic fragments and our dataset, showcasing similar results.

Assessment of mood after severe acquired brain injury: Interviews with UK clinical psychologists and medical professionals.

Mood assessment is challenging when people have cognitive and receptive communication impairments after severe brain injury. This study explored how UK-based medical and psychology professionals working with people with severe cognitive and communication impairments after brain injury assess mood in this population. Following their participation in an online survey, professionals were invited to participate in individual semi-structured interviews. Interviews were analysed using thematic analysis to label explicit data (semantic themes) and implicit data (latent themes). Twenty-three clinical psychologists and nine medical professionals participated in online or in-person interviews. Both groups explicitly reported using a combination of collateral information, history, observations, and patient interviews when assessing mood in this population. Medical professionals did not routinely use standardised measures and clinical psychologists often adjusted them when they used them. The respondents discussed difficulties conceptualising depression after brain injury, the experience needed by the assessor, and the need for an individualised approach for this population. Clinical psychologists discussed the pressures of working in healthcare systems and medical professionals discussed how symptoms may influence prescription choices. Seven latent themes were labelled which highlighted additional challenges and complexities experienced by those assessing mood, beyond the actual assessment process itself. No 'gold standard' approach to assessing mood in those with cognitive and communication difficulties after severe brain injury was identified. There was overlap in assessment approaches but no clear consensus. Interviewees felt that mood assessment must be approached differently in this population and that self-report measures are not useful.

Revealing the Unseen: AI Chain on LLMs for Predicting Implicit Data Flows to Generate Data Flow Graphs in Dynamically-Typed Code

Data flow graphs (DFGs) capture definitions (defs) and uses across program blocks, which is a fundamental program representation for program analysis, testing and maintenance. However, dynamically-typed programming languages like Python present implicit data flow issues that make it challenging to determine def-use flow information at compile time. Static analysis methods like Soot and WALA are inadequate for handling these issues, and manually enumerating comprehensive heuristic rules is impractical. Large pre-trained language models (LLMs) offer a potential solution, as they have powerful language understanding and pattern matching abilities, allowing them to predict implicit data flow by analyzing code context and relationships between variables, functions, and statements in code. We propose leveraging LLMs’ in-context learning ability to learn implicit rules and patterns from code representation and contextual information to solve implicit data flow problems. To further enhance the accuracy of LLMs, we design a five-step Chain of Thought (CoT) and break it down into an AI chain, with each step corresponding to a separate AI unit to generate accurate DFGs for Python code. Our approach’s performance is thoroughly assessed, demonstrating the effectiveness of each AI unit in the AI Chain. Compared to static analysis, our method achieves 82% higher def coverage and 58% higher use coverage in DFG generation on implicit data flow. We also prove the indispensability of each unit in the AI Chain. Overall, our approach offers a promising direction for building software engineering tools by utilizing foundation models, eliminating significant engineering and maintenance effort, but focusing on identifying problems for AI to solve.

Online grocery shopping recommender systems: Common approaches and practices

Food recommender systems have been developed for the online environment to support shoppers in making informed decisions. These systems analyze the extensive data collected to infer consumer preferences and needs, providing relevant product recommendations accordingly. Despite the potential of recommender systems as a strategic marketing tool in the online grocery shopping environment, there has been limited effort to systematically analyze approaches of prior studies on recommender systems for online grocery shoppers along the five stages of recommendation delivery: (1) identify recommendation goal, (2) acquire consumer data, (3) compute, (4) evaluate, and (5) present the recommendation. Therefore, this paper examines the advancements in each stage of delivering grocery recommendations to consumers from 2018 to March 2023. We performed a search strategy resulting in 50 papers dedicated to recommender systems for online grocery shoppers, which contrasts with previous research that typically examined recipe and meal recommendations that were merely meant to inspire users on what to cook. Findings reveal a prevalence of preference-based systems with limited integration of explicit consumer data, and often lacking consent for implicit data usage. While advanced deep neural network models are getting more attention in the literature, evaluation methods tend to be system-oriented, overlooking essential user feedback and the efficacy of general metrics. This systematic literature review underscores the necessity for consumer engagement in system and interface design, aiming for grocery recommendation systems that improve customer experience, by ensuring inclusivity and prioritizing user-centered design.

IRDA: Implicit data augmentation for deep imbalanced regression

Imbalanced data distributions are prevalent in real-world classification and regression tasks. Data augmentation is a commonly employed technique to mitigate this issue, with implicit methods gaining attention for their effectiveness and efficiency. However, implicit data augmentation methods have not been extensively explored in the context of regression tasks. To address this gap, we introduce IRDA, a novel learning method for regression that incorporates implicit data augmentation. Our approach includes developing a new augmentation strategy specifically tailored for deep imbalanced regression tasks, and a regression loss function that is suitable for real-world data with imbalanced label distributions and non-uniformly distributed features. We derive an easily computable surrogate loss and propose two implicit data augmentation algorithms, one incorporating meta-learning and one without. Additionally, we provide a regularization perspective to offer a deeper understanding of IRDA. We evaluate IRDA on five datasets, including a large-scale dataset, demonstrating its effectiveness in mitigating the adverse effects of imbalanced data distribution and its adaptability to various regression tasks.

Thematic Analysis of Using Visual Methods to Understand Healthcare Teams

In healthcare, teams are essential in solving today’s toughest challenges. However, narrow disciplinary perspectives have limited our current understanding of how teams function in various healthcare contexts. Researchers often rely on traditional survey methods as their primary tool, which can prevent them from gathering comprehensive data. To overcome this limitation, sharing stories and narratives visually allows participants to create representations of their perceptions of team experiences and complex encounters using a variety of approaches. Despite the potential of this method, there is little empirical guidance on how to use it in health sciences research. To address this gap and provide guidance on using visual methods, particularly for analysing team function, we contacted researchers worldwide specializing in qualitative research methods. These researchers had published experience using visual methods. Over a year, three virtual, asynchronous brainstorming sessions were conducted with 16 researchers. The brainstorming sessions consisted of respondents receiving a survey with open-ended prompts to gather insight into the best practices for using visual methods. Researchers agreed that visual methods can gather implicit data, break down researcher-participant power dynamics, and improve a study’s accessibility to under-represented groups. It was highlighted, when interpreting drawings, co-analysis with the creator should be facilitated to mitigate biases and support interpretation. This use of visual methods can aid the creation of tacit and more nuanced descriptions of complex phenomena and improve team function through a deeper understanding and respect of each members’ needs/experiences/perspectives. Future studies should pilot visual methods with different healthcare teams to further investigate team dynamics and how findings can be used to optimize team function.

Smooth-Guided Implicit Data Augmentation for Domain Generalization.

The training process of a domain generalization (DG) model involves utilizing one or more interrelated source domains to attain optimal performance on an unseen target domain. Existing DG methods often use auxiliary networks or require high computational costs to improve the model's generalization ability by incorporating a diverse set of source domains. In contrast, this work proposes a method called Smooth-Guided Implicit Data Augmentation (SGIDA) that operates in the feature space to capture the diversity of source domains. To amplify the model's generalization capacity, a distance metric learning (DML) loss function is incorporated. Additionally, rather than depending on deep features, the suggested approach employs logits produced from cross entropy (CE) losses with infinite augmentations. A theoretical analysis shows that logits are effective in estimating distances defined on original features, and the proposed approach is thoroughly analyzed to provide a better understanding of why logits are beneficial for DG. Moreover, to increase the diversity of the source domain, a sampling-based method called smooth is introduced to obtain semantic directions from interclass relations. The effectiveness of the proposed approach is demonstrated through extensive experiments on widely used DG, object detection, and remote sensing datasets, where it achieves significant improvements over existing state-of-the-art methods across various backbone networks.

Rasch model assessment of algebraic word problem among year 8 Malaysian students

<p>Word problems continue to be a challenge for students today. All students must meet the prerequisites for problem solving and reasoning skills, which are important components of the critical thinking component of 21st century skills. This study is being conducted to assess students’ strategies for solving word problems with numbers, consecutives, and ages. The Rasch model is used to analyze the item difficulty level of word problems and students’ strategies for solving ten-word problems at various levels of item difficulty in a similar trait. Then, Pearson correlation analysis is used to investigate the item difficulty level in relation to linguistic, algebraic, and arithmetic factors of word problems before evaluating students’ performance on solving these word problems using various strategies. Rasch model found these algebraic word-problem questions are slightly harder for year 8 Malaysian students in relative to an international standard. Meanwhile, the item difficulty of word problems is driven by linguistic and algebra factors where students can score accurately if the word problems contained explicit information. However, the students encountered difficulties while losing their solution strategy when the questions contained implicit data that demanded critical thinking ability.</p>

Fault Diagnosis of the Rolling Bearing by a Multi-Task Deep Learning Method Based on a Classifier Generative Adversarial Network.

Accurate fault diagnosis is essential for the safe operation of rotating machinery. Recently, traditional deep learning-based fault diagnosis have achieved promising results. However, most of these methods focus only on supervised learning and tend to use small convolution kernels non-effectively to extract features that are not controllable and have poor interpretability. To this end, this study proposes an innovative semi-supervised learning method for bearing fault diagnosis. Firstly, multi-scale dilated convolution squeeze-and-excitation residual blocks are designed to exact local and global features. Secondly, a classifier generative adversarial network is employed to achieve multi-task learning. Both unsupervised and supervised learning are performed simultaneously to improve the generalization ability. Finally, supervised learning is applied to fine-tune the final model, which can extract multi-scale features and be further improved by implicit data augmentation. Experiments on two datasets were carried out, and the results verified the superiority of the proposed method.

A novel complex-high-order graph convolutional network paradigm: ChyGCN

Abstract In recent years, there has been a growing interest in Graph Convolutional Networks (GCN). However, existing GCN and variants are predominantly based on simple graph or hypergraph structures, which restricts their ability to handle complex data correlations in practical applications. These limitations stem from the difficulty in establishing multiple hierarchies and acquiring adaptive weights for each of them. To address this issue, this paper introduces the latest concept of complex hypergraphs and constructs a versatile high-order multi-level data correlation model. This model is realized by establishing a three-tier structure of complexes-hypergraphs-vertices. Specifically, we start by establishing hyperedge clusters on the foundational network, utilizing a second-order hypergraph structure to depict potential correlations. For this second-order structure, truncation methods are applied to assess and generate a three-layer composite structure. During the construction of the composite structure, an adaptive learning strategy is implemented to merge correlations across different levels. We evaluate this model on several popular datasets and compare it with recent state-of-the-art methods. The comprehensive assessment results demonstrate that the proposed model surpasses existing methods, particularly in modeling implicit data correlations (the classification accuracy of nodes on five public datasets Cora, Citeseer, Pubmed, Github Web ML, and Facebook is 86.1±0.33, 79.2±0.35, 83.1±0.46, 83.8±0.23, and 80.1±0.37, respectively). This indicates that our approach holds advantages when handling datasets with implicit multi-level structures.

Investigating the Use of Deep Learning and Implicit Feedback in K12 Educational Recommender Systems

Recommender systems are increasingly being used in university or online education. However, recommender systems still have not found major usage in K12 education. This may be because of unique challenges that recommender systems face when used by a young and diverse population. Regardless, recommender systems for K12 education could provide many benefits for students and teachers, such as the simplification of personalized learning. Some of the issues with K12 educational recommender systems may be solved by the use of deep learning and implicit feedback. As such, we investigated the use of deep recommendation algorithms and implicit feedback for K12 educational recommender systems. To do this, we compared metrics for highly cited traditional and deep recommendation algorithms trained on explicit and implicit data. We found that recommendation algorithms using deep learning as a group do not differ in performance compared to traditional recommendation algorithms. We also found that the use of implicit feedback led to higher performance than using explicit feedback. The best performing algorithm used both deep learning and implicit feedback. We conclude that the deep learning can be a benefit for K12 recommender systems, particularly when the ordered sequence of items is carefully accounted for. We also conclude that researchers and developers must carefully consider which feedback contributes the most information for learning.

Kernelized Deep Learning for Matrix Factorization Recommendation System Using Explicit and Implicit Information.

In the current matrix factorization recommendation approaches, the item and the user latent factor vectors are with the same dimension. Thus, the linear dot product is used as the interactive function between the user and the item to predict the ratings. However, the relationship between real users and items is not entirely linear and the existing recommendation model of matrix factorization faces the challenge of data sparsity. To this end, we propose a kernelized deep neural network recommendation model in this article. First, we encode the explicit user-item rating matrix in the form of column vectors and project them to higher dimensions to facilitate the simulation of nonlinear user-item interaction for enhancing the connection between users and items. Second, the algorithm of association rules is used to mine the implicit relation between users and items, rather than simple feature extraction of users or items, for improving the recommendation performance when the datasets are sparse. Third, through the autoencoder and kernelized network processing, the implicit data are connected with the explicit data by the multilayer perceptron network for iterative training instead of doing simple linear weighted summation. Finally, the predicted rating is output through the hidden layer. Extensive experiments were conducted on four public datasets in comparison with several existing well-known methods. The experimental results indicated that our proposed method has obtained improved performance in data sparsity and prediction accuracy.

A Step Forward in Identifying Socially Desirable Respondents: An Integrated Machine Learning Model Considering T‐Scores, Response Time, Kinematic Indicators, and Eye Movements

Context: In high‐stakes assessments, such as court cases or managerial evaluations, decision‐makers heavily rely on psychological testing. These assessments often play a crucial role in determining important decisions that affect a person’s life and have a significant impact on society.Problem Statement: Research indicates that many psychological assessments are compromised by respondents’ deliberate distortions and inaccurate self‐presentations. Among these sources of bias, socially desirable responding (SDR) describes the tendency to provide overly positive self‐descriptions. This positive response bias can invalidate test results and lead to inaccurate assessments.Objectives: The present study is aimed at investigating the utility of mouse‐ and eye‐tracking technologies for detecting SDR in psychological assessments. By integrating these technologies, the study sought to develop more effective methods for identifying when respondents are presenting themselves in a favorable light.Methods: Eighty‐five participants completed the Lie (L) and Correction (K) scales of the Minnesota Multiphasic Personality Inventory‐2 (MMPI‐2) twice: once answering honestly and once presenting themselves in a favorable light, with the order of conditions balanced. Repeated measures univariate analyses were conducted on L and K scale T‐scores, as well as on mouse‐ and eye‐tracking features, to compare the honest and instructed SDR conditions. Additionally, machine learning models were developed to integrate T‐scores, kinematic indicators, and eye movements for predicting SDR.Results: The results showed that participants in the SDR condition recorded significantly higher T‐scores, longer response times, wider mouse trajectories, and avoided looking at the answers they intended to fake, compared to participants in the honest condition. Machine learning algorithms predicted SDR with 70%–78% accuracy.Conclusion: New assessment strategies using mouse‐ and eye‐tracking can help practitioners identify whether data is genuine or fabricated, potentially enhancing decision‐making accuracy.Implications: Combining self‐report measures with implicit data can improve SDR detection, particularly in managerial, organizational, and forensic contexts where precise assessments are crucial.

Variability across time in implicit weight-related bias: Random noise or meaningful fluctuations?

Social psychologists have struggled with the vexing problem of variability over time in implicit bias. While many treat such variability as unexplainable error, we posit that some temporal variability, whether within persons or across society at large, reflects meaningful and predictable fluctuation based on shifts in the social-cultural context. We first examined fluctuations at the group-level in a Project Implicit data set of female participants who completed the Weight Implicit Association Test between 2004 and 2018 (N = 259,613). Extending our prior work showing that mass media celebrity fat-shaming increased women's implicit antifat bias, we show that celebrity body positivity events reduced such bias (Study 1a). We then focused on a specific form of body positivity-that is, celebrity "push-back" in response to fat-shaming. Whereas fat-shaming without antibias push-back was associated with spikes in negative weight attitudes, fat-shaming with push-back showed no change in such bias (Study 1b). Critically, however, closer analysis revealed that this apparent stability was due to the canceling out of opposing negative (fat-shaming) and subsequent positive (body positivity) influences-an effect that was obscured when the window of observation was expanded. Finally, in Study 2, we examined parallel effects at the individual level in a daily diary study. Consistent with the group-level, between-subjects data, women's intraindividual fluctuations in implicit attitudes were reliably predicted based on prior-day exposure to fat-shaming and/or body positivity influences. Taken together, our work highlights how both group- and individual-level variability across time can be meaningfully explained rather than treated as unexplainable or left as unexplained. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

How being perceived to be an artist boosts feelings of attraction in others

Music production is a universal phenomenon reaching far back into our past. Given its ubiquity, evolution theorists have postulated adaptive functions for music, such as strengthening in-group cohesion, intimidating enemies, or promoting child bonding. Here, we focus on a longstanding Darwinian hypothesis, suggesting that music production evolved as a vehicle to display an individual’s biological fitness in courtship competition, thus rendering musicality a sexually selected trait. We also extend this idea to visual artists. In our design, we employed different versions of naturalistic portraits that manipulated the presence or absence of visual cues suggesting that the person was an artist or a non-artist (e.g., farmer, teacher, physician). Participants rated each portrayed person’s appeal on multiple scales, including attractiveness, interestingness, sympathy, and trustworthiness. Difference scores between portrait versions revealed the impact of the artistic/non-artistic visual cues. We thus tested Darwin’s hypothesis on both a within-subject and within-stimulus level. In addition to this implicit approach, we collected explicit ratings on the appeal of artists versus non-artists. The results demonstrate divergent findings for both types of data, with only the explicit statements corroborating Darwin’s hypothesis. We discuss this divergence in detail, along with the particular role of interestingness revealed by the implicit data.

Predictive Geochemical Exploration: Inferential Generation of Modern Geochemical Data, Anomaly Detection and Application to Northern Manitoba

Geochemical surveys contain an implicit data lifecycle or pipeline that consists of data generation (e.g., sampling and analysis), data management (e.g., quality assurance and control, curation, provisioning and stewardship) and data usage (e.g., mapping, modeling and hypothesis testing). The current integration of predictive analytics (e.g., artificial intelligence, machine learning, data modeling) into the geochemical survey data pipeline occurs almost entirely within the data usage stage. In this study, we predict elemental concentrations at the data generation stage and explore how predictive analytics can be integrated more thoroughly across the data lifecycle. Inferential data generation is used to modernize lake sediment geochemical data from northern Manitoba (Canada), with results and interpretations focused on elements that are included in the Canadian Critical Minerals list. The results are mapped, interpreted and used for downstream analysis through geochemical anomaly detection to locate further exploration targets. Our integration is novel because predictive modeling is integrated into the data generation and usage stages to increase the efficacy of geochemical surveys. The results further demonstrate how legacy geochemical data are a significant data asset that can be predictively modernized and used to support time-sensitive mineral exploration of critical minerals that were unanalyzed in original survey designs. In addition, this type of integration immediately creates the possibility of a new exploration framework, which we call predictive geochemical exploration. In effect, it eschews sequential, grid-based and fixed resolution sampling toward data-driven, multi-scale and more agile approaches. A key outcome is a natural categorization scheme of uncertainty associated with further survey or exploration targets, whether they are covered by existing training data in a spatial or multivariate sense or solely within the coverage of inferred secondary data. The uncertainty categorization creates an effective implementation pathway for future multi-scale exploration by focusing data generation activities to de-risk survey practices.

Automated multimodal sensemaking: Ontology-based integration of linguistic frames and visual data

Frame evocation from visual data is an essential process for multimodal sensemaking, due to the multimodal abstraction provided by frame semantics. However, there is a scarcity of data-driven approaches and tools to automate it. We propose a novel approach for explainable automated multimodal sensemaking by linking linguistic frames to their physical visual occurrences, using ontology-based knowledge engineering techniques. We pair the evocation of linguistic frames from text to visual data as “framal visual manifestations”. We present a deep ontological analysis of the implicit data model of the Visual Genome image dataset, and its formalization in the novel Visual Sense Ontology (VSO). To enhance the multimodal data from this dataset, we introduce a framal knowledge expansion pipeline that extracts and connects linguistic frames – including values and emotions – to images, using multiple linguistic resources for disambiguation. It then introduces the Visual Sense Knowledge Graph (VSKG), a novel resource. VSKG is a queryable knowledge graph that enhances the accessibility and comprehensibility of Visual Genome’s multimodal data, based on SPARQL queries. VSKG includes frame visual evocation data, enabling more advanced forms of explicit reasoning, analysis and sensemaking. Our work represents a significant advancement in the automation of frame evocation and multimodal sense-making, performed in a fully interpretable and transparent way, with potential applications in various fields, including the fields of knowledge representation, computer vision, and natural language processing.

Improving personalized medical implant design: An innovative and automated method leveraging implicit modeling

Advancements in manufacturing technologies have made the design and surgical applications of personalized implants increasingly popular and widespread. This study presents a fast, parametric, and automatic workflow for creating customized implants to be used in the surgical treatment of humerus bone traumas. The workflow operates on various humerus fractures, such as “Butterfly Fragment” fractures, taking into account patient-specific bone geometries and areas where screws will be applied. The workflow is carried out using implicit modelling and nTopology software. Initially, a replica of the “Butterfly Fragment” fracture is created for the study, and the damaged bone tissue is removed. Subsequently, a filler structure connecting the two main fragments is obtained using a porous structure. The workflow continues with the identification of areas where screws will be applied around the bone trauma and the selection of the surface where the implant will be created. Using selected points on the surface mesh and single-dimensional beams guided by these points, the external geometry and screw holes of the implant are shaped. The workflow is completed by offsetting certain structures created in intermediate steps and applying surface treatments to obtain the final shape and thickness of the implant, which is compatible with the bone structure. At the end of the process, the workflow is converted into a single function block, which can be shared for use by an implant designer. This enables the widespread application of personalized implant designs. The workflow accelerates the design process of the implant, providing implicit data suitable for production using additive manufacturing. This study offers a preparatory process for the rapid design and production of customized implants for patients. Future research may include comparisons of CAD design and implicit modelling methods, time studies, process modifications, optimizations, and investigations related to additive manufacturing printers. This study contributes to shortening the time until the application of surgical treatment methods by accelerating the design and production process of customized implants.

Criterion-based Heterogeneous Collaborative Filtering for Multi-behavior Implicit Recommendation

Recent years have witnessed the explosive growth of interaction behaviors in multimedia information systems, where multi-behavior recommender systems have received increasing attention by leveraging data from various auxiliary behaviors such as tip and collect. Among various multi-behavior recommendation methods, non-sampling methods have shown superiority over negative sampling methods. However, two observations are usually ignored in existing state-of-the-art non-sampling methods based on binary regression: (1) users have different preference strengths for different items, so they cannot be measured simply by binary implicit data; (2) the dependency across multiple behaviors varies for different users and items. To tackle the above issue, we propose a novel non-sampling learning framework named C riterion-guided H eterogeneous C ollaborative F iltering (CHCF). CHCF introduces both upper and lower thresholds to indicate selection criteria, which will guide user preference learning. Besides, CHCF integrates criterion learning and user preference learning into a unified framework, which can be trained jointly for the interaction prediction of the target behavior. We further theoretically demonstrate that the optimization of Collaborative Metric Learning can be approximately achieved by the CHCF learning framework in a non-sampling form effectively. Extensive experiments on three real-world datasets show the effectiveness of CHCF in heterogeneous scenarios.