We present LinkSAGE, an innovative framework that integrates Graph Neural Networks (GNNs) into large-scale personalized job matching systems, designed to address the complex dynamics of LinkedIn's extensive professional network. Our approach capitalizes on a novel job marketplace graph, the largest and most intricate of its kind in industry, with billions of nodes and edges. This graph is not merely extensive but also richly detailed, encompassing member and job nodes along with key attributes, thus creating an expansive and interwoven network. A key innovation in LinkSAGE is its training and serving methodology, which effectively combines inductive graph learning on a heterogeneous, evolving graph with an encoder-decoder GNN model. This methodology decouples the training of the GNN model from that of existing Deep Neural Network (DNN) models, eliminating the need for frequent GNN retraining while maintaining up-to-date graph signals in near real-time, allowing for the effective integration of GNN insights through transfer learning. The subsequent nearline inference system serves the GNN encoder within a real-world setting, significantly reducing online latency and obviating the need for costly real-time GNN infrastructure. Validated across multiple online A/B tests in diverse product scenarios, LinkSAGE demonstrates marked improvements in member engagement, relevance matching, and member retention, confirming its generalizability and practical impact.

Graph recommendation methods, representing a connected interaction perspective, reformulate user–item interactions as graphs to leverage graph structure and topology to recommend and have proved practical effectiveness at scale. Large language models (LLMs), representing a textual generative perspective, excel at modeling user languages, understanding behavioral contexts, capturing user–item semantic relationships, analyzing textual sentiments, and generating coherent and contextually relevant texts as recommendations. However, there is a gap between the connected graph perspective and the text generation perspective as the task formulations are different. A research question arises: how can we effectively integrate the two perspectives for more personalized RecSys? To fill this gap, we propose to incorporate graph-edge information into LLMs via prompt and attention innovations. We reformulate recommendations as a probabilistic generative problem using prompts. We develop a framework to incorporate graph edge information from the prompt and attention mechanisms for graph-structured LLM recommendations. We develop a new prompt design that brings in both first-order and second-order graph relationships; we devise an improved LLM attention mechanism to embed direct the spatial and connectivity information of edges. Our evaluation of real-world datasets demonstrates the framework’s ability to understand connectivity information in graph data and to improve the relevance and quality of recommendation results. Our code is released at: https://github.com/anord-wang/LLM4REC.git .

Job marketplace is a heterogeneous graph composed of interactions among members (job-seekers), companies, and jobs. Understanding and modeling job marketplace can benefit both job seekers and employers, ultimately contributing to the greater good of the society. However, existing graph neural network (GNN)-based methods have shallow understandings of the associated textual features and heterogeneous relations. To address the above challenges, we propose PLM4Job, a job marketplace foundation model that tightly couples pretrained language models (PLM) with job market graph, aiming to fully utilize the pretrained knowledge and reasoning ability to model member/job textual features as well as various member-job relations simultaneously. In the pretraining phase, we propose a heterogeneous ego-graph-based prompting strategy to model and aggregate member/job textual features based on the topological structure around the target member/job node, where entity type embeddings and graph positional embeddings are introduced accordingly to model different entities and their heterogeneous relations. Meanwhile, a proximity-aware attention alignment strategy is designed to dynamically adjust the attention of the PLM on ego-graph node tokens in the prompt, such that the attention can be better aligned with job marketplace semantics. Extensive experiments at LinkedIn demonstrate the effectiveness of PLM4Job.

Recently, there has been growing interest in developing the next-generation recommender systems (RSs) based on pretrained large language models (LLMs). However, the semantic gap between natural language and recommendation tasks is still not well addressed, leading to multiple issues such as spuriously correlated user/item descriptors, ineffective language modeling on user/item data, inefficient recommendations via auto-regression, etc. In this paper, we propose CLLM4Rec, the first generative RS that tightly integrates the LLM paradigm and ID paradigm of RSs, aiming to address the above challenges simultaneously. We first extend the vocabulary of pretrained LLMs with user/item ID tokens to faithfully model user/item collaborative and content semantics. Accordingly, a novel soft+hard prompting strategy is proposed to effectively learn user/item collaborative/content token embeddings via language modeling on RS-specific corpora, where each document is split into a prompt consisting of heterogeneous soft (user/item) tokens and hard (vocab) tokens and a main text consisting of homogeneous item tokens or vocab tokens to facilitate stable and effective language modeling. In addition, a novel mutual regularization strategy is introduced to encourage CLLM4Rec to capture recommendation-related information from noisy user/item content. Finally, we propose a novel recommendation-oriented finetuning strategy for CLLM4Rec, where an item prediction head with multinomial likelihood is added to the pretrained CLLM4Rec backbone to predict hold-out items based on soft+hard prompts established from masked user-item interaction history, where recommendations of multiple items can be generated efficiently without hallucination.

Recommender systems (RSs) have become an indispensable part of online platforms. With the growing concerns of algorithmic fairness, RSs are not only expected to deliver high-quality personalized content, but are also demanded not to discriminate against users based on their demographic information. However, existing RSs could capture undesirable correlations between sensitive features and observed user behaviors, leading to biased recommendations. Most fair RSs tackle this problem by completely blocking the influences of sensitive features on recommendations. But since sensitive features may also affect user interests in a fair manner (e.g., race on culture-based preferences), indiscriminately eliminating all the influences of sensitive features inevitably degenerate the recommendations quality and necessary diversities. To address this challenge, we propose a path-specific fair RS (PSF-RS) for recommendations. Specifically, we summarize all fair and unfair correlations between sensitive features and observed ratings into two latent proxy mediators, where the concept of path-specific bias (PS-Bias) is defined based on path-specific counterfactual inference. Inspired by Pearl's minimal change principle, we address the PS-Bias by minimally transforming the biased factual world into a hypothetically fair world, where a fair RS model can be learned accordingly by solving a constrained optimization problem. For the technical part, we propose a feasible implementation of PSF-RS, i.e., PSF-VAE, with weakly-supervised variational inference, which robustly infers the latent mediators such that unfairness can be mitigated while necessary recommendation diversities can be maximally preserved simultaneously. Experiments conducted on semi-simulated and real-world datasets demonstrate the effectiveness of PSF-RS.

Knowledge graphs (KGs) are of great importance in various artificial intelligence systems, such as question answering, relation extraction, and recommendation. Nevertheless, most real-world KGs are highly incomplete, with many missing relations between entities. To discover new triples (i.e., head entity, relation, tail entity), many KG completion algorithms have been proposed in recent years. However, a vast majority of existing studies often require a large number of training triples for each relation, which contradicts the fact that the frequency distribution of relations in KGs often follows a long tail distribution, meaning a majority of relations have only very few triples. Meanwhile, since most existing large-scale KGs are constructed automatically by extracting information from crowd-sourcing data using heuristic algorithms, plenty of errors could be inevitably incorporated due to the lack of human verification, which greatly reduces the performance for KG completion. To tackle the aforementioned issues, in this paper, we study a novel problem of error-aware few-shot KG completion and present a principled KG completion framework REFORM. Specifically, we formulate the problem under the few-shot learning framework, and our goal is to accumulate meta-knowledge across different meta-tasks and generalize the accumulated knowledge to the meta-test task for error-aware few-shot KG completion. To address the associated challenges resulting from insufficient training samples and inevitable errors, we propose three essential modules neighbor encoder, cross-relation aggregation, and error mitigation in each meta-task. Extensive experiments on three widely used KG datasets demonstrate the superiority of the proposed framework REFORM over competitive baseline methods.

Given a collection of items to display, such as news, videos, or products, how can we optimize their presentation order to maximize user engagements, such as click-through rate, viewing time, and the number of purchases? The problem becomes more complicated when the items are displayed in a grid-based, 2-dimensional presentation on a widescreen. For example, many E-Commerce websites such as Amazon and Etsy are displaying their products in a grid-like format, and so are streaming services like Youtube and Netflix. Unlike 1-dimensional space, where products can be naturally ranked in a vertical order, the presentation in 2-dimensional space poses a novel challenge about how to find the best presentation order - should we put the best listing on the top left corner, or the central position on the second row? We are aware that many traditional methods can be applied to solve the problem, such as conducting an attention heatmap web test, or a randomization experiment by shuffling positions of listings. However, both tests are costly to perform and they may downgrade the quality of users' search experience. By contrast, we focus on utilizing existing search log data to reveal propensity of positions, which is readily available and ubiquitously abundant.

A two-sided travel marketplace is an E-Commerce platform where users can both host tours or activities and book them as a guest. When a new guest visits the platform, given tens of thousands of available listings, a natural question is that what kind of activities or trips are the best fit. In order to answer the question, a recommender system needs to both understand characteristics of its inventories, and to know the preferences of each individual guest. In this work, we present our efforts on building a recommender system for Airbnb Experiences, a two-sided online marketplace for tours and activities. Traditional recommender systems rely on abundant user-listing interactions. Airbnb Experiences is an emerging business where many listings and guests are new to the platform. Instead of passively waiting for data to accumulate, we propose novel approaches to identify key features of a listing and estimate guest preference with limited data availability.

As opposed to manual feature engineering which is tedious and difficult to scale, network embedding has attracted a surge of research interests as it automates the feature learning on graphs. The learned low-dimensional node vectors ease the knowledge discovery on graphs by enabling various off-the-shelf machine learning tools to be directly applied. Recent research has shown that the past decade of network embedding approaches either explicitly factorize a carefully designed matrix or are closely related to implicit matrix factorization, with the fundamental assumption that the factorized node connectivity matrix is low-rank. Nonetheless, the global low-rank assumption does not necessarily hold especially when the factorized matrix encodes complex node interactions, and the resultant single low-rank embedding matrix is insufficient to capture all the observed connectivity patterns. In this regard, we propose a novel multi-level network embedding framework BoostNE, which can learn multiple node embeddings of different granularity from coarse to fine without imposing the prevalent global low-rank assumption. The proposed BoostNE method is also in line with the successful gradient boosting method in ensemble learning. We demonstrate the superiority of the proposed BoostNE framework by comparing it with existing state-of-the-art network embedding methods on various datasets.

Relational learning has been proposed to cope with the interdependency among linked instances in a network, and it is a fundamental tool to categorize social network users for various tasks. However, the emerging widespread of misinformation in social networks, information that is inaccurate or false, poses novel challenges to utilizing social media data. Malicious users may actively manipulate their content and characteristics, which easily lead to a noisy dataset. Hence, it is intricate for traditional relational learning approaches to deliver an accurate predictive model in the presence of misinformation. In this work, we precisely focus on the problem by proposing a joint framework that simultaneously constructs a relational learning model and mitigates the effect of misinformation by restraining anomalous points. Empirical results on real-world social media data prove the superiority of the proposed approach, Relational Learning with Misinformation (RLM), over traditional approaches on modeling social network users.

In recent years, product search engines have emerged as a key factor for online businesses. According to a recent survey, over 55% of online customers begin their online shopping journey by searching on an E-Commerce (EC) website like Amazon as opposed to a generic web search engine like Google. Information retrieval research to date has been focused on optimizing search ranking algorithms for web documents while little attention has been paid to product search. There are several intrinsic differences between web search and product search that make the direct application of traditional search ranking algorithms to EC search platforms difficult. First, the success of web and product search is measured differently; one seeks to optimize for relevance while the other must optimize for both relevance and revenue. Second, when using real-world EC transaction data, there is no access to manually annotated labels. In this paper, we address these differences with a novel learning framework for EC product search called LETORIF (LEarning TO Rank with Implicit Feedback). In this framework, we utilize implicit user feedback signals (such as user clicks and purchases) and jointly model the different stages of the shopping journey to optimize for EC sales revenue. We conduct experiments on real-world EC transaction data and introduce a a new evaluation metric to estimate expected revenue after re-ranking. Experimental results show that LETORIF outperforms top competitors in improving purchase rates and total revenue earned.

Link prediction targets to predict the future node interactions mainly based on the current network snapshot. It is a key step in understanding the formation and evolution of the underlying networks; and has practical implications in many real-world applications, ranging from friendship recommendation, click through prediction to targeted advertising. Most existing efforts are devoted to plain networks and assume the availability of network structure in memory before link prediction takes place. However, this assumption is untenable as many real-world networks are affiliated with rich node attributes, and often, the network structure and node attributes are both dynamically evolving at an unprecedented rate. Even though recent studies show that node attributes have an added value to network structure for accurate link prediction, it still remains a daunting task to support link prediction in an online fashion on such dynamic attributed networks. As changes in the dynamic attributed networks are often transient and can be endless, link prediction algorithms need to be efficient by making only one pass of the data with limited memory overhead. To tackle these challenges, we study a novel problem of streaming link prediction on dynamic attributed networks and present a novel framework - SLIDE. Methodologically, SLIDE maintains and updates a low-rank sketching matrix to summarize all observed data, and we further leverage the sketching matrix to infer missing links on the fly. The whole procedure is theoretically guaranteed, and empirical experiments on real-world dynamic attributed networks validate the effectiveness and efficiency of the proposed framework.

When a message, such as a piece of news, spreads in social networks, how can we classify it into categories of interests, such as genuine or fake news? Classification of social media content is a fundamental task for social media mining, and most existing methods regard it as a text categorization problem and mainly focus on using content features, such as words and hashtags. However, for many emerging applications like fake news and rumor detection, it is very challenging, if not impossible, to identify useful features from content. For example, intentional spreaders of fake news may manipulate the content to make it look like real news. To address this problem, this paper concentrates on modeling the propagation of messages in a social network. Specifically, we propose a novel approach, TraceMiner, to (1) infer embeddings of social media users with social network structures; and (2) utilize an LSTM-RNN to represent and classify propagation pathways of a message. Since content information is sparse and noisy on social media, adopting TraceMiner allows to provide a high degree of classification accuracy even in the absence of content information. Experimental results on real-world datasets show the superiority over state-of-the-art approaches on the task of fake news detection and news categorization.

Relational learning exploits relationships among instances manifested in a network to improve the predictive performance of many network mining tasks. Due to its empirical success, it has been widely applied in myriad domains. In many cases, individuals in a network are highly idiosyncratic. They not only connect to each other with a composite of factors but also are often described by some content information of high dimensionality specific to each individual. For example in social media, as user interests are quite diverse and personal; posts by different users could differ significantly. Moreover, social content of users is often of high dimensionality which may negatively degrade the learning performance. Therefore, it would be more appealing to tailor the prediction for each individual while alleviating the issue related to the curse of dimensionality. In this paper, we study a novel problem of Personalized Relational Learning and propose a principled framework PRL to personalize the prediction for each individual in a network. Specifically, we perform personalized feature selection and employ a small subset of discriminative features customized for each individual and some common features shared by all to build a predictive model. On this account, the proposed personalized model is more human interpretable. Experiments on real-world datasets show the superiority of the proposed PRL framework over traditional relational learning methods.

The explosive use of social media, in information dissemination and communication, has also made it a popular platform for the spread of rumors. Rumors could be easily propagated and received by a large number of users in social media, resulting in catastrophic effects in the physical world in a very short period. It is a challenging task, if not impossible, to apply classical supervised learning methods to the early detection of rumors, since the labeling process is time-consuming and labor-intensive. Motivated by the fact that abundant label information of historical rumors is publicly available, in this paper, we propose to investigate whether knowledge learned from historical data could potentially help identify newly emerging rumors. In particular, since a disputed factual claim arouses certain reactions such as curiosity, skepticism, and astonishment, we identify and utilize patterns from prior labeled data to help reveal emergent rumors. Experimental results on real-world data sets demonstrate the effectiveness. Further experiments are conducted to show how much earlier it can detect an emerging rumor than traditional approaches.

The presence of bots has been felt in many aspects of social media. Twitter, one example of social media, has especially felt the impact, with bots accounting for a large portion of its users. These bots have been used for malicious tasks such as spreading false information about political candidates and inflating the perceived popularity of celebrities. Furthermore, these bots can change the results of common analyses performed on social media. It is important that researchers and practitioners have tools in their arsenal to remove them. Approaches exist to remove bots, however they focus on precision to evaluate their model at the cost of recall. This means that while these approaches are almost always correct in the bots they delete, they ultimately delete very few, thus many bots remain. We propose a model which increases the recall in detecting bots, allowing a researcher to delete more bots. We evaluate our model on two real-world social media datasets and show that our detection algorithm removes more bots from a dataset than current approaches.

The automatic and early detection of rumors is of paramount importance as the spread of information with questionable veracity can have devastating consequences. This became starkly apparent when, in early 2013, a compromised Associated Press account issued a tweet claiming that there had been an explosion at the White House. This tweet resulted in a significant drop for the Dow Jones Industrial Average. Most existing work in rumor detection leverages conversation statistics and propagation patterns, however, such patterns tend to emerge slowly requiring a conversation to have a significant number of interactions in order to become eligible for classification. In this work, we propose a method for classifying conversations within their formative stages as well as improving accuracy within mature conversations through the discovery of implicit linkages between conversation fragments. In our experiments, we show that current state-of-the-art rumor classification methods can leverage implicit links to significantly improve the ability to properly classify emergent conversations when very little conversation data is available. Adopting this technique allows rumor detection methods to continue to provide a high degree of classification accuracy on emergent conversations with as few as a single tweet. This improvement virtually eliminates the delay of conversation growth inherent in current rumor classification methods while significantly increasing the number of conversations considered viable for classification.

Feature selection has shown its effectiveness to prepare high-dimensional data for many data mining and machine learning tasks. Traditional feature selection algorithms are mainly based on the assumption that data instances are independent and identically distributed. However, this assumption is invalid in networked data since instances are not only associated with high dimensional features but also inherently interconnected with each other. In addition, obtaining label information for networked data is time consuming and labor intensive. Without label information to direct feature selection, it is difficult to assess the feature relevance. In contrast to the scarce label information, link information in networks are abundant and could help select relevant features. However, most networked data has a lot of noisy links, resulting in the feature selection algorithms to be less effective. To address the above mentioned issues, we propose a robust unsupervised feature selection framework NetFS for networked data, which embeds the latent representation learning into feature selection. Therefore, content information is able to help mitigate the negative effects from noisy links in learning latent representations, while good latent representations in turn can contribute to extract more meaningful features. In other words, both phases could cooperate and boost each other. Experimental results on real-world datasets demonstrate the effectiveness of the proposed framework.

Relational learning has been proposed to cope with the interdependency among linked instances in social network analysis, which often adopts network connectivity and social media content for prediction. A common assumption in existing relational learning methods is that data instances are equally important. The algorithms developed based on the assumption may be significantly affected by outlier data and thus less robust. In the meantime, it has been well established in social sciences that actors are naturally of different social status in a social network. Motivated by findings from social sciences, in this paper, we investigate whether social status analysis could facilitate relational learning. Particularly, we propose a novel framework RESA to model social status using the network structure. It extracts robust and intrinsic latent social dimensions for social actors, which are further exploited as features for supervised learning models. The proposed method is applicable for real-world relational learning problems where noise exists. Extensive experiments are conducted on datasets obtained from real-world social media platforms. Empirical results demonstrate the effectiveness of RESA and further experiments are conducted to help understand the effects of parameter settings to the proposed model and how local social status works.