Modern neural networks transform vast datasets into continuous embedding spaces, translating semantic relationships into geometric structures. The key to unlocking their full potential lies in making these representations interpretable. This talk presents a simple theory underlining the interpretability of the embedding space and how this principle can allow us to analyze the data, design new metrics, and model the dynamics of the system, moving beyond black-box models to data-driven interpretable insights.

Quantifying identity fusion-the psychological merging of self with another entity or abstract target (e.g., a religious group, political party, ideology, value, brand, belief, etc.)-is vital for understanding a wide range of group-based human behaviors.We introduce the Cognitive Linguistic Identity Fusion Score (CLIFS), a novel metric that integrates cognitive linguistics with large language models (LLMs), which builds on implicit metaphor detection.Unlike traditional pictorial and verbal scales, which require controlled surveys or direct field contact, CLIFS delivers fully automated, scalable assessments while maintaining strong alignment with the established verbal measure.In benchmarks, CLIFS outperforms both existing automated approaches and human annotation.As a proof of concept, we apply CLIFS to violence risk assessment to demonstrate that it can improve violence risk assessment by more than 240%.Building on our identification of a new NLP task and early success, we underscore the need to develop larger, more diverse datasets that encompass additional fusion-target domains and cultural backgrounds to enhance generalizability and further advance this emerging area.CLIFS models and code are public at

Link prediction---a task of distinguishing actual hidden edges from random unconnected node pairs---is one of the quintessential tasks in graph machine learning. Despite being widely accepted as a universal benchmark and a downstream task for representation learning, the link prediction benchmark's validity has rarely been questioned. Here, we show that the common edge sampling procedure in the link prediction task has an implicit bias toward high-degree nodes. This produces a highly skewed evaluation that favors methods overly dependent on node degree. In fact a ``null'' link prediction method based solely on node degree can yield nearly optimal performance in this setting. We propose a degree-corrected link prediction benchmark that offers a more reasonable assessment and better aligns with the performance on the recommendation task. Finally, we demonstrate that the degree-corrected benchmark can more effectively train graph machine-learning models by reducing overfitting to node degrees and facilitating the learning of relevant structures in graphs.

Narratives are foundation of human cognition and decision making. Because narratives play a crucial role in societal discourses and spread of misinformation and because of the pervasive use of social media, the narrative dynamics on social media can have profound societal impact. Yet, systematic and computational understanding of online narratives faces critical challenge of the scale and dynamics; how can we reliably and automatically extract narratives from massive amount of texts? How do narratives emerge, spread, and die? Here, we propose a systematic narrative discovery framework that fill this gap by combining change point detection, semantic role labeling (SRL), and automatic aggregation of narrative fragments into narrative networks. We evaluate our model with synthetic and empirical data — two Twitter corpora about COVID-19 and 2017 French Election. Results demonstrate that our approach can recover major narrative shifts that correspond to the major events.

1

How does our society appreciate the uniqueness of cultural products? This fundamental puzzle has intrigued scholars in many fields, including psychology, sociology, anthropology, and marketing. It has been theorized that cultural products that balance familiarity and novelty are more likely to become popular. However, a cultural product's novelty is typically multifaceted. This paper uses songs as a case study to study the multiple facets of uniqueness and their relationship with success. We first unpack the multiple facets of a song's novelty or uniqueness and, next, measure its impact on a song's popularity. We employ a series of statistical models to study the relationship between a song's popularity and novelty associated with its lyrics, chord progressions, or audio properties. Our analyses performed on a dataset of over fifty thousand songs find a consistently negative association between all types of song novelty and popularity. Overall we found a song's lyrics uniqueness to have the most significant association with its popularity. However, audio uniqueness was the strongest predictor of a song's popularity, conditional on the song's genre. We further found the theme and repetitiveness of a song's lyrics to mediate the relationship between the song's popularity and novelty. Broadly, our results contradict the "optimal distinctiveness theory'' (balance between novelty and familiarity) and call for an investigation into the multiple dimensions along which a cultural product's uniqueness could manifest.

Graph embedding maps a graph into a convenient vector-space representation for graph analysis and machine learning applications. Many graph embedding methods hinge on a sampling of context nodes based on random walks. However, random walks can be a biased sampler due to the structural properties of graphs. Most notably, random walks are biased by the degree of each node, where a node is sampled proportionally to its degree. The implication of such biases has not been clear, particularly in the context of graph representation learning. Here, we investigate the impact of the random walks' bias on graph embedding and propose residual2vec, a general graph embedding method that can debias various structural biases in graphs by using random graphs. We demonstrate that this debiasing not only improves link prediction and clustering performance but also allows us to explicitly model salient structural properties in graph embedding.

Abstract Collaborations are pervasive in current science. Collaborations have been studied and encouraged in many disciplines. However, little is known about how a team really functions from the detailed division of labor within. In this research, we investigate the patterns of scientific collaboration and division of labor within individual scholarly articles by analyzing their co‐contributorship networks. Co‐contributorship networks are constructed by performing the one‐mode projection of the author–task bipartite networks obtained from 138,787 articles published in PLoS journals. Given an article, we define 3 types of contributors: Specialists, Team‐players, and Versatiles. Specialists are those who contribute to all their tasks alone; team‐players are those who contribute to every task with other collaborators; and versatiles are those who do both. We find that team‐players are the majority and they tend to contribute to the 5 most common tasks as expected, such as “data analysis” and “performing experiments.” The specialists and versatiles are more prevalent than expected by our designed 2 null models. Versatiles tend to be senior authors associated with funding and supervision. Specialists are associated with 2 contrasting roles: the supervising role as team leaders or marginal and specialized contributors.

Because word semantics can substantially change across communities and contexts, capturing domain-specific word semantics is an important challenge.Here, we propose SEMAXIS, a simple yet powerful framework to characterize word semantics using many semantic axes in wordvector spaces beyond sentiment.We demonstrate that SEMAXIS can capture nuanced semantic representations in multiple online communities.We also show that, when the sentiment axis is examined, SEMAXIS outperforms the state-of-theart approaches in building domain-specific sentiment lexicons.

Clustering is a central approach for unsupervised learning. After clustering is applied, the most fundamental analysis is to quantitatively compare clusterings. Such comparisons are crucial for the evaluation of clustering methods as well as other tasks such as consensus clustering. It is often argued that, in order to establish a baseline, clustering similarity should be assessed in the context of a random ensemble of clusterings. The prevailing assumption for the random clustering ensemble is the permutation model in which the number and sizes of clusters are fixed. However, this assumption does not necessarily hold in practice; for example, multiple runs of K-means clustering reurns clusterings with a fixed number of clusters, while the cluster size distribution varies greatly. Here, we derive corrected variants of two clustering similarity measures (the Rand index and Mutual Information) in the context of two random clustering ensembles in which the number and sizes of clusters vary. In addition, we study the impact of one-sided comparisons in the scenario with a reference clustering. The consequences of different random models are illustrated using synthetic examples, handwriting recognition, and gene expression data. We demonstrate that the choice of random model can have a drastic impact on the ranking of similar clustering pairs, and the evaluation of a clustering method with respect to a random baseline; thus, the choice of random clustering model should be carefully justified.

Online social network providers have become treasure troves of information for marketers and researchers. To profit from their data while honoring the privacy of their customers, social networking services share `anonymized' social network datasets, where, for example, identities of users are removed from the social network graph. However, by using external information such as a reference social graph (from the same network or another network with similar users), researchers have shown how such datasets can be de-anonymized. These approaches use `network alignment' techniques to map nodes from the reference graph into the anonymized graph and are often sensitive to larger network sizes, the number of seeds, and noise --- which may be added to preserve privacy. We propose a divide-and-conquer approach to strengthen the power of such algorithms. Our approach partitions the networks into `communities' and performs a two-stage mapping: first at the community level, and then for the entire network. Through extensive simulation on real-world social network datasets, we show how such community-aware network alignment improves de-anonymization performance under high levels of noise, large network sizes, and a low number of seeds. Even when nodes cannot be explicitly mapped, the community structure can be mapped between both networks, thus reducing the anonymity of users. For example, for our (real-world) Twitter dataset with 90,000 nodes, 20% noise, and 16 seeds, the state-of-the-art technique reduces anonymity by 0 bits, whereas our approach reduces anonymity by 9.71 bits (with 40% of nodes mapped).

The modular organization of networks of individual neurons interwoven through synapses has not been fully explored due to the incredible complexity of the connectivity architecture. Here we use the modularity-based community detection method for directed, weighted networks to examine hierarchically organized modules in the complete wiring diagram (connectome) of Caenorhabditis elegans (C. elegans) and to investigate their topological properties. Incorporating bilateral symmetry of the network as an important cue for proper cluster assignment, we identified anatomical clusters in the C. elegans connectome, including a body-spanning cluster, which correspond to experimentally identified functional circuits. Moreover, the hierarchical organization of the five clusters explains the systemic cooperation (e.g., mechanosensation, chemosensation, and navigation) that occurs among the structurally segregated biological circuits to produce higher-order complex behaviors.

Online social networking services are among the most popular Internet services according to Alexa.com and have become a key feature in many Internet services. Users interact through various features of online social networking services: making friend relationships, sharing their photos, and writing comments. These friend relationships are expected to become a key to many other features in web services, such as recommendation engines, security measures, online search, and personalization issues. However, we have very limited knowledge on how much interaction actually takes place over friend relationships declared online. A friend relationship only marks the beginning of online interaction.

User Generated Content (UGC) is re-shaping the way people watch video and TV, with millions of video producers and consumers. In particular, UGC sites are creating new viewing patterns and social interactions, empowering users to be more creative, and developing new business opportunities. To better understand the impact of UGC systems, we have analyzed YouTube, the world's largest UGC VoD system. Based on a large amount of data collected, we provide an in-depth study of YouTube and other similar UGC systems. In particular, we study the popularity life-cycle of videos, the intrinsic statistical properties of requests and their relationship with video age, and the level of content aliasing or of illegal content in the system. We also provide insights on the potential for more efficient UGC VoD systems (e.g. utilizing P2P techniques or making better use of caching). Finally, we discuss the opportunities to leverage the latent demand for niche videos that are not reached today due to information filtering effects or other system scarcity distortions. Overall, we believe that the results presented in this paper are crucial in understanding UGC systems and can provide valuable information to ISPs, site administrators, and content owners with major commercial and technical implications.

Social networking services are a fast-growing business in the Internet. However, it is unknown if online relationships and their growth patterns are the same as in real-life social networks. In this paper, we compare the structures of three online social networking services: Cyworld, MySpace, and orkut, each with more than 10 million users, respectively. We have access to complete data of Cyworld's ilchon (friend) relationships and analyze its degree distribution, clustering property, degree correlation, and evolution over time. We also use Cyworld data to evaluate the validity of snowball sampling method, which we use to crawl and obtain partial network topologies of MySpace and orkut. Cyworld, the oldest of the three, demonstrates a changing scaling behavior over time in degree distribution. The latest Cyworld data's degree distribution exhibits a multi-scaling behavior, while those of MySpace and orkut have simple scaling behaviors with different exponents. Very interestingly, each of the two e ponents corresponds to the different segments in Cyworld's degree distribution. Certain online social networking services encourage online activities that cannot be easily copied in real life; we show that they deviate from close-knit online social networks which show a similar degree correlation pattern to real-life social networks.