Data Mining And Its Applications For Micro Entities In Internet Finance

Internet Finance is an emerging Internet-based financial mechanism by which the traditional financial institutions or Internet enterprises could achieve the funding,pay-ment,investment,or intermediary service online.Internet Finance is one of the most important mechanism in modern Fintech.Compared with the traditional finance,In-ternet Finance has some distinct advantages,such as high efficiency,low cost,wide range and convenient operating.However,on the other hand,with the prevalence of Internet Finance,the barrier of financial market is getting lower,the liquidity of users is increasing,the supervision and management are becoming more difficult.Recent years have witnessed the rapid development of Internet Finance platforms which have been becoming more and more popular.However,the market volatility and service complexity of Internet Finance bring new challenges to the research paradigms in traditional economics and finance.Fortunately,large-scale user behavior data and transaction data in these platforms are accumulated,which also provide opportunities for researchers to explore the potential values in Internet financial environments,solving the problems in Internet financial markets,and developing the data-driven intelligent financial services.Actually,the research on Internet Financial scenario has received wide attention in both the academic and industrial circles.Based on this background,this paper sys-tematically.carried out a series of exploratory research on data mining methods and applications for the micro entities in Internet Finance.Specifically,according to the three main roles(users,products,market management)in the Internet financial market,this research work is carried out from these three aspects.Along the line of micro users,we propose the investment recommendation with risk management,and the portfolio selection from the multi-objective perspective;along the line of products,we propose to track the funding dynamics with a hierarchical time series model,and a sequential approach to market state modeling based on Bayesian Hidden Markov model;along the line of market management,we propose a joint deep survival model to jointly model on donation recurrence and donor retention.The main efforts and contributions of this paper can be summarized as follows.First,we provide a comprehensive survey on the research Internet Finance,to the best of our knowledge,which is the first focused effort in this field.Internet Finance is the emerging research hot topic in both the academic and industrial circles.We firstprovide a systematic taxonomy for Internet Finance by summarizing different types of mainstream platforms and comparing their working mechanisms in detail.Then,we review and organize the recent advances on this area from various perspectives.Also,we propose our opinions on the prospects and suggest some future research directions in this field.In the meantime,for constructing the data-driven research,we collect sev-eral large-scale datasets from different representative platforms and thoroughly explore these data.Second,from the aspect of micro users,we propose the investment recommenda-tion with risk management,the portfolio selection from the multi-objective perspective and the Nash equilibrium-based group recommendation.For most users,it is very diffi-cult for them to make decision on investment especially when they are facing thousands of investing products.For that,we propose the investment recommendation with risk management.Specifically,we focus on the personalized investment recommendation by reconstructing the two steps for investment decision making:"what to buy" and"how much money to pay".We first profile the users and items via their Bipartite In-vestment Network and generate a candidate investment list for each investor that tackles"what to buy" problem.Furthermore,we optimize the share of each recommended can-didate by incorporating the investments an investor currently holds,thus solving the"how much money to pay" problem.Finally,extensive experimental results on a large-scale real-world dataset show the effectiveness of our model under various evaluation metrics.The trading mechanisms of different Internet platforms are different.Therefore,users will have more complex considerations in their investment decisions.Currently,many platforms adopt the auction-based trading rule.Thus,rational users often pursue multiple objectives,e.g.,non-default probability,fully-funded probability and winning-bid probability.Also,they have the portfolio perspective in their mind,i.e.,usually selecting more than one products(i.e.,a portfolio)to bid in each investment.To that end,we present a holistic study on portfolio selections.Specifically,we first propose to adopt gradient boosting decision tree,which combines both static features and dynamic features,to assess products from multiple objectives.Then,we propose two strategies,i.e.,weighted objective optimization strategy and multi-objective optimization strategy,to select portfolios for users.Finally,extensive experiments on a large-scale real-world dataset demonstrate the effectiveness of our solutions.Thirdly,from the aspect of products,we propose to track the funding dynamics with a hierarchical time series model,and a sequential approach to market state model-ing based on Bayesian Hidden Markov model.In these financial platforms,the dynam-ics,i.e.,daily funding amount on campaigns and perks,are the most concerned issue for creators and platforms.As the most popular type of crowdfunding,a campaign of-ten sets several types of "rewards"(i.e.,perks)with different prices,thus,participants could make different monetary contributions by selecting different perks.A special goal is to forecast the funding amount for a given campaign and its perks in the future days.Specifically,we formalize the dynamics in crowdfunding as a hierarchical time series,i.e.,campaign level and perk level.Specific to each level,we develop a special regression by modeling the decision making process of the crowd(visitors and back-ing probability)and exploring various factors that impact the decision;on this basis,an enhanced switching regression is proposed at each level to address the heterogeneity of funding sequences.Further,we employ a revision matrix to combine the two-level base forecasts for the final forecasting.We conduct extensive experiments on a real-world crowdfunding data collected from Indiegogo.The experimental results clearly demon-strate the effectiveness of our approaches on tracking the dynamics in crowdfunding.Actually,the dynamics of funding are determined by the hidden market state of products.For mining the relations between the observable funding dynamics and the hidden market state,we propose a sequential approach to market state modeling based on Bayesian Hidden Markov model.Specifically,we first propose two enhanced se-quential models by extending the Bayesian Hidden Markov model(BHMM),namely listing-BHMM(L-BHMM)and listing and marketing-BHMM(LM-BHMM),for learn-ing the latent semantics between products' market states and users' bidding behaviors.Particularly,L-BHMM is a straightforward model that only considers the local observa-tions of a product itself,while LM-BHMM considers not only the product information but also the global information of current market(e.g.,the competitive and comple-mentary relations among products).Finally,we construct extensive experiments on two real-world datasets,which clearly validate the effectiveness of our models in terms of relevant applications.Finally,from the aspect of market management,we propose a joint deep survival model to jointly model the donation recurrence and donor retention.In Internet Finance,the liquidity of users is serious.Especially because the non-profit nature,the situation of donor retention for donation-based crowdfunding as well as traditional charities is extremely serious.Thus,analyzing the factors of and further predicting the donor's behaviors are very urgent.In this study,we focus on two types of behavioral events,e.g.,donation recurrence and donor retention.Specifically,we propose a joint deep sur-vival model,which can integrate heterogeneous features,e.g.,donor motives,recently donated projects,social contacts,to jointly model the donation recurrence and donor retention since these two types of behavioral events are highly relevant.Additionally,we model the censoring phenomenon and dependence relations of different behaviors from the survival analysis view by designing multiple innovative constraints and in-corporating them into the objective functions.Finally,we conduct extensive analysis and validation experiments with the large-scale data.The experimental results clearly demonstrate the effectiveness of our proposed models for analyzing and predicting the donation recurrence and donor retention in crowdfunding.