Projects

The diversity of opinions is both cure and curse for the effective use of crowd-sourced intelligence. To unify crowd-sourced intelligence for a sensible decision, we develop an algorithmic approach for decision aggregation by properly quantifying reliability of information from multiple sources. The key idea is to model reliability propagation of decisions on an ensemble of relevance graphs, where the optimizations of reliability propagation and graph ensemble are mutually reinforced. The propagated reliability can be used to aggregate intelligence from multiple sources and facilitate decision-making by leveraging various types of inter-correlations of information sources as well as information subjects. Meanwhile, the optimized graph ensemble can retain the relevance structures with respect to the crowd-sourced intelligence. We show that, when aggregating analysts' recommendations in stock markets, our approach is effective and superior compared to alternative methods.

Extracting typical career paths from large-scale and unstructured talent profiles has recently attracted increasing research attention. However, various challenges arise in effectively analyzing self-reported career records. Inspired by recent advances in neural networks and embedding models, we develop a novel career path clustering approach with two major components. First, we formulate an embedded Markov framework to learn job embeddings from longitudinal career records and further use them to compute dynamic embeddings of career paths. Second, to cope with heterogeneous career path clusters, we estimate a mixture of Markov models to optimize cluster-wise job embeddings with a prior embedded space shared by multiple clusters. We conduct extensive experiments with our framework to investigate its algorithmic performance and extract meaningful clustering patterns of career paths in the information technology (IT) industry. The results show that our approach can naturally discover distinct sequential path clusters and reveal valuable insights for career path characterization.

Gaining highly skilled human capital is one of the primary reasons for corporate mergers and acquisitions (M&A), especially for knowledge intensive industry. However, the inevitable tensions brought by the divergent cultures and organizational misalignment during the M&A process result in high talent turnover rate and ultimately the integration failure. Hence, it is imperative to understand and prepare for the potential effects of M&A process on the employee turnover. To this end, we propose a novel dual-fit model induced heterogeneous Graph Neural Network (GNN) model to predict the talent turnover trend in the post-M&A process, by taking into account the complex relationship among the acquirer firm, the acquiree firm, and the acquired employees. Specifically, we creatively design a dual-fit model comprised of both the firm-level compatibility and employeefirm fit. Extensive evaluations on large-scale real-world data clearly demonstrate the effectiveness of our approach.

It is critical and important for venture investors to find high-potential startups at their early stages. Indeed, many efforts have been made to study the key factors for the success of startups through the topological analysis of the heterogeneous information network of people, startup, and venture firms or representation learning of latent startup profile features. However, the existing topological analysis lacks an in-depth understanding of heterogeneous information. Also, the approach based on representation learning heavily relies on domain-specific knowledge for feature selections. Instead, in this paper, we propose a Scalable Heterogeneous Graph Markov Neural Network (SHGMNN) for identifying the high-potential startups. The general idea is to use graph neural networks (GNN) to learn effective startup representations through end-to-end efficient training and model the label dependency among startups through Maximum A Posterior (MAP) inference. Specifically, we first define different metapaths to capture various semantics over the heterogeneous information network (HIN) and aggregate all semantic information into a summated graph structure. To predict the high-potential early-stage startups, we introduce GNN to diffuse the information over the summated graph. We then adopt a MAP inference over Hinge-Loss Markov Random Fields to enforce label dependency. Here, a pseudolikelihood variational expectation maximization (EM) framework is incorporated to optimize both MAP inference and GNN iteratively: The E-step calculates the inference, and the M-step updates the GNN. For efficiency concerns, we develop a GNN with a lightweight linear diffusion architecture to perform graph propagation over web-scale heterogeneous information networks. Finally, extensive experiments and case studies on real-world datasets demonstrate the superiority of SHGMNN.

Extensive efforts have been made by both academics and practitioners to understand inter-firm competitive relationship owing to its profound impacts on multiple key business goals, e.g., company benchmarking, marketing strategy planning, and talent acquisition. However, it has never been an easy task to fully characterize firms and assess the competitive relationship among them mainly due to the challenge of information heterogeneity. In this regard, we propose a novel IT artifact for firm profiling and inter-firm competition assessment guided by Information System Design Theory (ISDT). We start by constructing a Heterogeneous Occupation Network (HON) using employees’ occupation details and education attainments. Then we adopt a Methpath2Vec-based heterogeneous network embedding model to learn firms’ latent profiles (embeddings). Using the firm embeddings as input, we train multiple supervised classifiers to assess the competitive relationship among the firms. Following the logic of design as a search process, we demonstrate the utility of our IT artifact with extensive experimental study and in-depth discussions. Our study also reveals that employees’ occupation and education information significantly contribute to the identification of the focal firm’s potential competitors.

Venture capital investments play a powerful role in fueling the emergence and growth of early-stage startups. However, only a small fraction of venture-backed startups can survive and exit successfully. Prior data-driven prediction-based or recommendation-based solutions are incapable of providing effective and actionable strategies on proper investment timing and amounts for startups across different investment rounds. In this paper, we develop a novel reinforcement learning-based method, AlphaVC, to facilitate venture capitalists’ decision-making. Our policy-based RL agents can dynamically identify the best candidates and sequentially place the optimal investment amounts at proper rounds to maximize financial returns for a given portfolio. We retrieve company demographics and investment activity data from Crunchbase and test our model in two industry sectors: Financial Services and Information Technology. Our methodology demonstrates its efficacy and superiority in both ranking and portfolio-based performance metrics in comparison with various SOTA baseline methods.