Strong Homophily Research Articles

Exploring the role of edge distribution in graph convolutional networks

Graph Convolutional Networks (GCNs) have shown remarkable performance in processing graph-structured data by leveraging neighborhood information for node representation learning. While most GCN models assume strong homophily within the networks they handle, some models can also handle heterophilous graphs. However, the selection of neighbors participating in the node representation learning process can significantly impact these models’ performance. To address this, we investigate the influence of neighbor selection on GCN performance, focusing on the analysis of edge distribution through theoretical and empirical approaches. Based on our findings, we propose a novel GCN model called Graph Convolution Network with Improved Edge Distribution (GCN-IED). GCN-IED incorporates both direct edges, which rely on local neighborhood similarity, and hidden edges, obtained by aggregating information from multi-hop neighbors. We extensively evaluate GCN-IED on diverse graph benchmark datasets and observe its superior performance compared to other state-of-the-art GCN methods on heterophilous datasets. Our GCN-IED model, which considers the role of neighbors and optimizes edge distribution, provides valuable insights for enhancing graph representation learning and achieving superior performance on heterophilous graphs.

Taming over-smoothing representation on heterophilic graphs

Graph Neural Networks (GNNs) have shown expressive modeling ability on graphs. An inherent assumption is that those connected graph nodes exhibit strong homophily, passing consistent label information into the associated central node representation. However, heterophilic graphs may hold inconsistent labels for those connected nodes, degenerating the performance of the typical message passing and exacerbating its over-smoothing issue. In this paper, to tame heterophilic graphs for GNNs, a novel label estimation-based message passing scheme is presented, which can further relieve the over-smoothing and lead the learned representations to be more distinguishable on graphs with either homophily or heterophily. Specifically, we optimize the edge weights of the graph by invoking the label estimation with the goal of returning a more effective adjacency matrix. With this matrix, a novel message passing scheme is presented, aggregating neighbor information with consistent labels to their central nodes. Meanwhile, a symmetric cross entropy is also employed to mitigate negative effects from noisy labels. With such guarantees in our new message passing, a uniform framework termed Label Estimation-based GNN (LE-GNN) is finally established. Extensive experiments demonstrate the ability of our model to prevent over-smoothing and show its effectiveness for node classification on both homophilic and heterophilic graphs.

Diffusion in small worlds with homophily and social reinforcement: A theoretical model

We introduce homophily in a percolation model of word-of-mouth diffusion in social networks by reorganizing the nodes according to similarity in preferences for adoption of an innovation. Such preferences are described by a “minimum utility requirement” for an agent to adopt. We show that homophily removes the non-linear relation between preferences and diffusion in the standard percolation model with a high diffusion regime (“hit”) and a low diffusion regime (“flop”). Instead, in a model with perfect homophily, the final diffusion scales linearly with individual preferences: all agents who are willing to adopt, do adopt the innovation. We also investigate the combined effect of homophily and social reinforcement in diffusion. Results indicate that social reinforcement renders clustered networks more efficient in terms of diffusion size for network with strong homophily, while the opposite is true for networks without homophily. The simple structure of our model allows to disentangle the effect of social influence, homophily and the network structure on diffusion. However, the controllability of the theoretical structure comes at the expenses of the realism of the model. For this, we discuss possible extensions and empirical applications.

Social networks of oncology clinicians as a means for increasing survivorship clinic referral

BackgroundSpecialized cancer survivorship clinics are recommended for addressing treatment-related health concerns of long-term survivors, but their relative newness in medical oncology necessitates strategies to expand services and clinic referrals. This study used social network analysis to identify personal and/or network factors associated with referral of patients to a survivorship clinic.MethodsWe conducted a cross-sectional social network survey of clinical personnel at a National Cancer Institute-designated comprehensive cancer center. Participants identified colleagues with whom they consult for advice (advice network) and/or discuss patient care (discussion network). Exponential random graph models and logistic regression were used to identify key opinion leaders in the network and factors associated with referral of patients to the center’s survivorship clinic.ResultsHere we show that of the respondents (n = 69), 78.0% report being aware of the survivorship clinic, yet only 30.4% had ever referred patients to it. Individuals tend to associate with others in the same occupational role (homophily). In the discussion network, holding an influential network position (betweenness centrality) is associated with patient referral to the survivorship clinic. In the advice network, several social workers are identified as opinion leaders.ConclusionsThis study shows that there is strong homophily in both networks, potentially inhibiting information sharing between groups. In designing an inclusive network intervention, persons occupying influential network positions and opinion leaders (e.g., social workers in this case) are well-positioned to promote survivorship clinic referrals.

Graph Pointer Neural Networks

Graph Neural Networks (GNNs) have shown advantages in various graph-based applications. Most existing GNNs assume strong homophily of graph structure and apply permutation-invariant local aggregation of neighbors to learn a representation for each node. However, they fail to generalize to heterophilic graphs, where most neighboring nodes have different labels or features, and the relevant nodes are distant. Few recent studies attempt to address this problem by combining multiple hops of hidden representations of central nodes (i.e., multi-hop-based approaches) or sorting the neighboring nodes based on attention scores (i.e., ranking-based approaches). As a result, these approaches have some apparent limitations. On the one hand, multi-hop-based approaches do not explicitly distinguish relevant nodes from a large number of multi-hop neighborhoods, leading to a severe over-smoothing problem. On the other hand, ranking-based models do not joint-optimize node ranking with end tasks and result in sub-optimal solutions. In this work, we present Graph Pointer Neural Networks (GPNN) to tackle the challenges mentioned above. We leverage a pointer network to select the most relevant nodes from a large amount of multi-hop neighborhoods, which constructs an ordered sequence according to the relationship with the central node. 1D convolution is then applied to extract high-level features from the node sequence. The pointer-network-based ranker in GPNN is joint-optimized with other parts in an end-to-end manner. Extensive experiments are conducted on six public node classification datasets with heterophilic graphs. The results show that GPNN significantly improves the classification performance of state-of-the-art methods. In addition, analyses also reveal the privilege of the proposed GPNN in filtering out irrelevant neighbors and reducing over-smoothing.

Herd immunity and epidemic size in networks with vaccination homophily.

We study how the herd immunity threshold and the expected epidemic size depend on homophily with respect to vaccine adoption. We find that the presence of homophily considerably increases the critical vaccine coverage needed for herd immunity and that strong homophily can push the threshold entirely out of reach. The epidemic size monotonically increases as a function of homophily strength for a perfect vaccine, while it is maximized at a nontrivial level of homophily when the vaccine efficacy is limited. Our results highlight the importance of vaccination homophily in epidemic modeling.

Epidemic spreading and digital contact tracing: Effects of heterogeneous mixing and quarantine failures.

Contact tracing via digital tracking applications installed on mobile phones is an important tool for controlling epidemic spreading. Its effectivity can be quantified by modifying the standard methodology for analyzing percolation and connectivity of contact networks. We apply this framework to networks with varying degree distributions, numbers of application users, and probabilities of quarantine failures. Further, we study structured populations with homophily and heterophily and the possibility of degree-targeted application distribution. Our results are based on a combination of explicit simulations and mean-field analysis. They indicate that there can be major differences in the epidemic size and epidemic probabilities which are equivalent in the normal susceptible-infectious-recovered (SIR) processes. Further, degree heterogeneity is seen to be especially important for the epidemic threshold but not as much for the epidemic size. The probability that tracing leads to quarantines is not as important as the application adoption rate. Finally, both strong homophily and especially heterophily with regard to application adoption can be detrimental. Overall, epidemic dynamics are very sensitive to all of the parameter values we tested out, which makes the problem of estimating the effect of digital contact tracing an inherently multidimensional problem.

Extremely Low Reciprocity and Strong Homophily in the World Largest MSM Social Network

Traditional survey-based methods are limited in the sample size and inference ability for the study of men who have sex with men (MSM), one of the most vulnerable groups at increased risk of HIV. Internet data, on the other hand, have provided publicly accessible information about such groups at unprecedented scale and resolution. Here we present statistics for the user demographics characteristics in the world's largest MSM geosocial networking application, Blued, and analyze the social network structure with 11,408,872 nodes and 838,910,078 edges extracted from user-following relationships on Blued. Network features, such as degree distribution, reciprocity, degree assortativity, homophily, and community are studied. We find that, in contrast to earlier analyses on social networks of general populations, the MSM social network is disassortative and shows extremely low reciprocity. Users in their twenties are excessively followed by users from all age groups, and network homophily for age and country are strong.

Graph Neural Networks with Heterophily

Graph Neural Networks (GNNs) have proven to be useful for many different practical applications. However, many existing GNN models have implicitly assumed homophily among the nodes connected in the graph, and therefore have largely overlooked the important setting of heterophily, where most connected nodes are from different classes. In this work, we propose a novel framework called CPGNN that generalizes GNNs for graphs with either homophily or heterophily. The proposed framework incorporates an interpretable compatibility matrix for modeling the heterophily or homophily level in the graph, which can be learned in an end-to-end fashion, enabling it to go beyond the assumption of strong homophily. Theoretically, we show that replacing the compatibility matrix in our framework with the identity (which represents pure homophily) reduces to GCN. Our extensive experiments demonstrate the effectiveness of our approach in more realistic and challenging experimental settings with significantly less training data compared to previous works: CPGNN variants achieve state-of-the-art results in heterophily settings with or without contextual node features, while maintaining comparable performance in homophily settings.

Close contact infection dynamics over time: insights from a second large-scale social contact survey in Flanders, Belgium, in 2010-2011

BackgroundIn 2010-2011, we conducted a social contact survey in Flanders, Belgium, aimed at improving and extending the design of the first social contact survey conducted in Belgium in 2006. This second social contact survey aimed to enable, for the first time, the estimation of social mixing patterns for an age range of 0 to 99 years and the investigation of whether contact rates remain stable over this 5-year time period.MethodsDifferent data mining techniques are used to explore the data, and the age-specific number of social contacts and the age-specific contact rates are modelled using a generalized additive models for location, scale and shape (GAMLSS) model. We compare different matrices using assortativeness measures. The relative change in the basic reproduction number (R0) and the ratio of relative incidences with 95% bootstrap confidence intervals (BCI) are employed to investigate and quantify the impact on epidemic spread due to differences in sex, day of the week, holiday vs. regular periods and changes in mixing patterns over the 5-year time gap between the 2006 and 2010-2011 surveys. Finally, we compare the fit of the contact matrices in 2006 and 2010-2011 to Varicella serological data.ResultsAll estimated contact patterns featured strong homophily in age and sex, especially for small children and adolescents. A 30% (95% BCI [17%; 37%]) and 29% (95% BCI [14%; 40%]) reduction in R0 was observed for weekend versus weekdays and for holiday versus regular periods, respectively. Significantly more interactions between people aged 60+ years and their grandchildren were observed on holiday and weekend days than on regular weekdays. Comparing contact patterns using different methods did not show any substantial differences over the 5-year time period under study.ConclusionsThe second social contact survey in Flanders, Belgium, endorses the findings of its 2006 predecessor and adds important information on the social mixing patterns of people older than 60 years of age. Based on this analysis, the mixing patterns of people older than 60 years exhibit considerable heterogeneity, and overall, the comparison of the two surveys shows that social contact rates can be assumed stable in Flanders over a time span of 5 years.

Consumer reactions to nudity in print advertising: Comparing same‐gender and opposite‐gender effects

AbstractIt is often assumed that exposure to nude stimuli in advertising influences consumer behavior positively. However, the empirical evidence concerning the effects of nudity on consumer reactions is inconclusive. The goal of this study is to disentangle the effects of opposite‐gender and same‐gender nudity on female and male consumers' reactions. This study, thereby, offers a framework for the appropriate choice of seminude or fully clothed human stimuli based on advertisers' objectives and consumer–model gender interactions. The empirical data were derived from a large‐scale market research initiative from Germany with 61,399 consumer evaluations of 147 real ads from 16 product categories. Female consumers show positive same‐gender results for both seminude and fully clothed female models, indicating strong homophily, but nonsignificant opposite‐gender effects on information search, positive attitude change, integration of brand into consideration set, and purchase intentions. In contrast, male consumers demonstrate a significantly positive and equally strong influence of opposite‐gender and same‐gender seminude stimuli on all of the four variables.

Homophily and peer-consumer behaviour in a peer-to-peer accommodation sharing economy platform

ABSTRACT Peer-to-peer (P2P) accommodation is considered as an important subsector of the ‘sharing economy’. P2P sharing economy (SE) platforms bring mostly individual actors called ‘peers’ together to create value. In general, interactions among peers in P2P SE platforms are considered novel and social with communal feelings such as friendliness and social satisfaction (Perren and Kozinets 2018, Lateral Exchange Markets: How Social Platforms Operate in a Networked Economy) and some research attention has been paid to peer interaction dynamics on P2P accommodation platforms in recent years. This study employs homophily, a potential user interaction dynamic but rarely studied in the context of SE platforms, and investigates its impact on peer-consumer behaviour in a P2P accommodation platform. We also examine homophily drivers in the context of accommodation sharing platforms. Our results suggest that homophily does contribute to peer-users’ consumption intention through trust and attitude, and that ethnicity and gender, two strong homophily factors identified in the literature, do not have statistically significant effects on homophily in this context. However, qualitative analysis of open-ended questions suggests that other homophily factors may be more influential in the accommodation sharing economy context. We believe that this study makes a meaningful contribution by expanding our understanding of peer-users’ interaction dynamics in the context of accommodation sharing platforms.

Global halal food discourse on social media: a text mining approach

ABSTRACT No previous studies have explored the distinct role played by halal food discourse on social media. This study fills this research gap by investigating the structure, dynamics and sentiments related to halal food tweets using methods borrowed from text mining and social network analytics. Based on sample of 11,700 world-wide random tweets, we used both the NRC and the AFINN lexicons to gauge worldwide sentiments towards halal food. Although results show a generally positive sentiment, the tweets also reflect a general concern about animal welfare. Community detection techniques revealed a strong homophily effect among different halal food communities. This effect occurs when actors in an environment similar to a virtual room deal with topics of common interest or discuss common agenda. Our findings hold important implications for different stakeholders as they underscore the relational nature of halal food networks and the importance of the strategic management of social media as a critical communication tool.

Abstract C015: Social network analysis with respondent driven sampling data: A study of ethnic and geographic integration on Guam and Hawaii

Abstract US Pacific Islanders are one of the fastest-growing population groups in the US, and serious disparities in health outcomes exist in this group, with cancer being the leading cause of death. Limited knowledge about the cancer communication practices and risk behaviors of US Pacific Islanders, as well as their social network structure and integration, makes effective cancer prevention interventions difficult, and contributes to disparities. In 2017, we conducted a survey on health communication on Guam and Hawaii using the respondent-driven sampling method (RDS) to recruit participants (N=733 for Hawaii, and N=533 for Guam). RDS is a network-based sampling technique where initial seed respondents recruit others from their social networks. The recruiting process repeats iteratively, thereby forming long referral chains. RDS provides a viable data for making inferences about the underlying network structure. During the sampling process two pieces of information are gathered. First, each recruiter-recruit dyad is documented. Second, respondents are asked how many other members of the target population they know and interact with. In this study, we used this information to create recruitment chains and test the quantity and quality of cross-gender, cross-ethnicity, cross-generation ties. We also analyzed the geographic distribution and spatial clustering of the samples. Self-reported network size (degree) was used to estimate of a homophily, a measure of the strength of association to one's own group beyond random mixing. In general, strong homophily was found for different ethnic groups (with values ranging from 0.7 to 0.8), whereas gender and age had weak homophily. Even though the participants tended to recruit spatially proximal peers, both samples in Guam and Hawaii eventually covered the whole catchment area. Citation Format: Grazyna Badowski, Louis Dulana, Lilnabeth P. Somera, Kevin Cassel, Hye-ryeon Lee, Janessa Roman, Jayson Morales. Social network analysis with respondent driven sampling data: A study of ethnic and geographic integration on Guam and Hawaii [abstract]. In: Proceedings of the Eleventh AACR Conference on the Science of Cancer Health Disparities in Racial/Ethnic Minorities and the Medically Underserved; 2018 Nov 2-5; New Orleans, LA. Philadelphia (PA): AACR; Cancer Epidemiol Biomarkers Prev 2020;29(6 Suppl):Abstract nr C015.

Model-based Respondent-driven sampling analysis for HIV prevalence in brazilian MSM

Respondent Driven Sampling study (RDS) is a population sampling method developed to study hard-to-reach populations. A sample is obtained by chain-referral recruitment in a network of contacts within the population of interest. Such self-selected samples are not representative of the target population and require weighing observations to reduce estimation bias. Recently, the Network Model-Assisted (NMA) method was described to compute the required weights. The NMA method relies on modeling the underlying contact network in the population where the RDS was conducted, in agreement with directly observable characteristics of the sample such as the number of contacts, but also with more difficult-to-measure characteristics such as homophily or differential characteristics according to the response variable. Here we investigated the use of the NMA method to estimate HIV prevalence from RDS data when information on homophily is limited. We show that an iterative procedure based on the NMA approach allows unbiased estimations even in the case of strong population homophily and differential activity and limits bias in case of preferential recruitment. We applied the methods to determine HIV prevalence in men having sex with men in Brazilian cities and confirmed a high prevalence of HIV in these populations from 3.8% to 22.1%.

Network Evolution of a Large Online MSM Dating Community: 2005–2018

Due to multiple sexual partners and low rates of condom use, the HIV infection rate among MSM (men who have sex with men) is much higher than that of the general population. In order to analyze the characteristics of online activities of MSM, and to understand the evolution of their social networks, in this study we collect a comprehensive dataset, covering the period from January 2005 to June 2018, from the largest Chinese online community, Baidu Tieba. We build an online dating network for MSM-related individuals in the gay-bar community, and analyze the network from static and dynamic aspects. It is found that there is a strong homophily regarding the cities where users reside when developing interactions with others, and that most network measurements tend to be stable at the later stages of evolution, while the size of the largest community fluctuates. This is an indication that the network is formed of rapidly flexible interactions which changes quickly. In comparison with studies on heterosexual networks, we find that the MSM dating network shows differences in many aspects, such as the positive degree-degree correlation and high clustering coefficient, suggesting different thinking and measures should be taken in the policy making of public health management towards the MSM population.

Social studies of scholarly life with sensor-based ethnographic observations

Social network analysis is playing an increasingly important role in sociological studies. At the same time, new technologies such as wearable sensors make it possible to collect new types of social network data. We employed RFID tags to capture face-to-face interactions of participants of two consecutive Ph.D. retreats of a graduate school on climate research. We use this data in order to explore how it may support ethnographic observations and to gain further insights on scholarly interactions. The unique feature of the data is the opportunity to distinguish short and long conversations, which often have a different nature from a sociological point of view. Furthermore, an advantage of this data is the availability of socio-demographic, research-related, and situational attributes of participants. We show that, even though an interaction partner is often found rather randomly during coffee breaks of retreats, a strong homophily between participants from the same institutions or research areas exists. We identify cores of the networks and participants who play ambassador roles between communities, e.g., persons who visit the retreat for the second time are more likely to be ambassadors. Overall, we show the usefulness and potential of RFID tags for scientometric studies.

Academic performance and behavioral patterns

Identifying the factors that influence academic performance is an essential part of educational research. Previous studies have documented the importance of personality traits, class attendance, and social network structure. Because most of these analyses were based on a single behavioral aspect and/or small sample sizes, there is currently no quantification of the interplay of these factors. Here, we study the academic performance among a cohort of 538 undergraduate students forming a single, densely connected social network. Our work is based on data collected using smartphones, which the students used as their primary phones for two years. The availability of multi-channel data from a single population allows us to directly compare the explanatory power of individual and social characteristics. We find that the most informative indicators of performance are based on social ties and that network indicators result in better model performance than individual characteristics (including both personality and class attendance). We confirm earlier findings that class attendance is the most important predictor among individual characteristics. Finally, our results suggest the presence of strong homophily and/or peer effects among university students.

A decade of dynamics of residential location, car ownership, activity, travel and land use in the Seattle metropolitan region

Using data of attitudes in the Puget Sound region we first identify predispositions in favor of car use (car loving persons and households) and distinguish them from other predispositions favoring other modes (transit and carsharing lovers). Then we explore if like-minded people (homophily) live together and examine heterogeneity within their households. To do this we analyze attitudinal data using multilevel latent class clustering that allows identification of groups of persons and groups of households jointly. Our analysis identifies distinct groups of people with different attitudes towards modes and we do not find strong homophily in attitudes within households. The distinction of attitudes among persons transfers well to the household level giving us the opportunity to identify and test differences among the different attitudinal groups using longitudinal records. We then move to a longitudinal analysis using a small sample of households that participated in more than ten years of the Puget Sound panel survey to explore the sequence of their residential location characteristics, car ownership, and travel. We find in descriptive statistics and in a second application of multilevel cluster models that carpool and transit loyalty persists over time by a portion of car pool and transit lovers. We also find sustained use of cars by car loving households, sustained car use of a neutral in attitudes group, a group of younger households with positive attitudes towards carsharing and sustained car sharing over time, and a possibly disenfranchised older household group lacking access to opportunities. The inclusion in the analysis of the two residential location characteristics (evolution of density and diversity around the household residence) enabled a more complete analysis and a clearer description of household context for both attitudes and behavior.

Formation of homophily in academic performance: Students change their friends rather than performance.

Homophily, the tendency of individuals to associate with others who share similar traits, has been identified as a major driving force in the formation and evolution of social ties. In many cases, it is not clear if homophily is the result of a socialization process, where individuals change their traits according to the dominance of that trait in their local social networks, or if it results from a selection process, in which individuals reshape their social networks so that their traits match those in the new environment. Here we demonstrate the detailed temporal formation of strong homophily in academic achievements of high school and university students. We analyze a unique dataset that contains information about the detailed time evolution of a friendship network of 6,000 students across 42 months. Combining the evolving social network data with the time series of the academic performance (GPA) of individual students, we show that academic homophily is a result of selection: students prefer to gradually reorganize their social networks according to their performance levels, rather than adapting their performance to the level of their local group. We find no signs for a pull effect, where a social environment of good performers motivates bad students to improve their performance. We are able to understand the underlying dynamics of grades and networks with a simple model. The lack of a social pull effect in classical educational settings could have important implications for the understanding of the observed persistence of segregation, inequality and social immobility in societies.