A multi-agent legal recommender system

Sedentary lifestyles and bad eating habits influence the onset of many serious health problems. Healthy behavior change is an arduous task, and requires a careful planning. In this thesis, we propose that behavior recommenders can help their users achieve healthy behavior change. Such a system should inspire its users with small, incremental and achievable goals. For this, it must resolve a trade-off between two opposing objectives: help the user achieve a steady improvement in target behavior, and avoid extreme goals that may injure or discourage the user. This is an unprecedented challenge in the recommender systems research. If the system understands the impacts of past interventions for behavior change, it can determine its usersâ behavioral responses to its own recommendations. This implies a specific data curation, in which we not only measure people's behavior but also deliberately introduce an intervention to monitor its effect on people's patterns. In turn, the system can use these existing users' information to derive the right procedure for effective recommendations. In this study we capitalize on this insight and develop InspiRE - our behavior recommender framework. Through InspiRE we propose the following contributions: 1) We design the data curation. 2) We develop the novel approaches for behavior profiling 3) We develop an evaluation process for this novel type of recommender system, and also compare it with traditional, similarity-based recommendation approach. We curate a dataset that contains information of daily step counts and social intervention for 83 people. InspiRE successfully uses the observations from this dataset, and proposes recommendations that are both effective and feasible. We also show that InspiRE can generalize to other dimensions of well being: we demonstrate this through a dataset that contains the snacking patterns of 73 people, who receive message-based interventions. We observe that InspiRE's recommendation strategy is in line with theories of behavior change.

Recommender system is the system which, by using automated information filtering technology, recommends products or services to the customers who are likely to be interested in. Those systems are widely used in many different Web retailers such as Amazon.com, Netfix.com, and CDNow.com. Various recommender systems have been developed. Among them, Collaborative Filtering (CF) has been known as the most successful and commonly used approach. CF identifies customers whose tastes are similar to those of a given customer, and recommends items those customers have liked in the past. Numerous CF algorithms have been developed to increase the performance of recommender systems. However, the relative performances of CF algorithms are known to be domain and data dependent. It is very time-consuming and expensive to implement and launce a CF recommender system, and also the system unsuited for the given domain provides customers with poor quality recommendations that make them easily annoyed. Therefore, predicting in advance whether the performance of CF recommender system is acceptable or not is practically important and needed. In this study, we propose a decision making guideline which helps decide whether CF is adoptable for a given application with certain transaction data characteristics. Several previous studies reported that sparsity, gray sheep, cold-start, coverage, and serendipity could affect the performance of CF, but the theoretical and empirical justification of such factors is lacking. Recently there are many studies paying attention to Social Network Analysis (SNA) as a method to analyze social relationships among people. SNA is a method to measure and visualize the linkage structure and status focusing on interaction among objects within communication group. CF analyzes the similarity among previous ratings or purchases of each customer, finds the relationships among the customers who have similarities, and then uses the relationships for recommendations. Thus CF can be modeled as a social network in which customers are nodes and purchase relationships between customers are links. Under the assumption that SNA could facilitate an exploration of the topological properties of the network structure that are implicit in transaction data for CF recommendations, we focus on density, clustering coefficient, and centralization which are ones of the most commonly used measures to capture topological properties of the social network structure. While network density, expressed as a proportion of the maximum possible number of links, captures the density of the whole network, the clustering coefficient captures the degree to which the overall network contains localized pockets of dense connectivity. Centralization reflects the extent to which connections are concentrated in a small number of nodes rather than distributed equally among all nodes. We explore how these SNA measures affect the performance of CF performance and how they interact to each other. Our experiments used sales transaction data from H department store, one of the well?known department stores in Korea. Total 396 data set were sampled to construct various types of social networks. The dependant variable measuring process consists of three steps; analysis of customer similarities, construction of a social network, and analysis of social network patterns. We used UCINET 6.0 for SNA. The experiments conducted the 3-way ANOVA which employs three SNA measures as dependant variables, and the recommendation accuracy measured by F1-measure as an independent variable. The experiments report that 1) each of three SNA measures affects the recommendation accuracy, 2) the density's effect to the performance overrides those of clustering coefficient and centralization (i.e., CF adoption is not a good decision if the density is low), and 3) however though the density is low, the performance of CF is comparatively good when the clustering coefficient is low. We expect that these experiment results help firms decide whether CF recommender system is adoptable for their business domain with certain transaction data characteristics.

I. ABSTRACT In Internet of Things (IoT), selling smart physical objects together with a compatible mobile app becomes an upcoming trend. The app allows to control or monitor the physical object and its sensors in an easy, ubiquitous, and user friendly way. In this post, we present a novel recommender system which uses publicly available data about these apps as a source for personalization. The proposed system infers user’s physical objects from exploring the installed apps on her mobile devices like smartphones and tablets and builds a digital inventory of physical objects of each user. Then, these inventories can be used to create personalized recommendations. A. Background Smartphones have become our daily companions and support us in almost any situation of our daily life. We use them for communication, gaming, news reading, watching movies, payment, mobile banking and many others. The way we use smartphones is correlated with individuals’ needs, interests, habits, and personality [1]. Custom-tailored content like recommendations, advertisement, personalized prices and search results can be presented to consumers, based on observed activities on their devices [2]. From the perspective of IoT, smartphones are becoming increasingly important because they are used as hubs between the physical objects and Internet. For instance, smartphones provide dashboards to manage our smart home appliance (like heating system, light-bulb, key, TV) as well as retrieving and presenting data from our smart gadgets (like pedometers, body sensors). Since there is no method to automatically get a list of individuals’ physical objects, offering fitting recommendations in IoT is difficult. Recent studies propose to gather data about installed apps on smartphones as a way to get valuable user information and to derive user profiles [3]. We believe that the installed apps are also a door opener for gathering information about the connected physical objects and we are able to automatically

We demonstrate how crowdsourcing can be used to automatically build a personalized tourist attraction recommender system, which tailors recommendations to specific individuals, so different people who use the system each get their own list of recommendations, appropriate to their own traits. Recommender systems crucially depend on the availability of reliable and large scale data that allows predicting how a new individual is likely to rate items from the catalog of possible items to recommend. We show how to automate the process of generating this data using crowdsourcing, so that such a system can be built even when such a dataset is not initially available. We first find possible tourist attractions to recommend by scraping such information from Wikipedia. Next, we use crowdsourced workers to filter the data, then provide their opinions regarding these items. Finally, we use machine learning methods to predict how new individuals are likely to rate each attraction, and recommend the items with the highest predicted ratings.

A multi-agent legal recommender system

Multi-agent systems exhibit unexpected, emergent behavior as a result of the complexity of agent behaviors and their interactions. Despite significant research interest in the past decades, computational methods to identify and analyze emergence as it happens are still needed. This paper proposes a software architecture for identifying emergent behavior in a multi-agent system as it happens, using interval-based snapshots and emergent behavior metrics. We propose various distance functions to compare between the multi-agent system under analysis and systems that have been previously shown to exhibit emergent behavior.

Solving complex but structured problems in a decentralized manner via multiagent collaboration has received much attention in recent years. This is natural, as on one hand, multiagent systems usually possess a structure that determines the allowable interactions among the agents; and on the other hand, the single most pressing need in a cooperative multiagent system is to coordinate the local policies of autonomous agents with restricted capabilities to serve a system-wide goal. The presence of uncertainty makes this even more challenging, as the agents face the additional need to learn the unknown environment parameters while forming (and following) local policies in an online fashion. In this paper, we provide the first Bayesian reinforcement learning (BRL) approach for distributed coordination and learning in a cooperative multiagent system by devising two solutions to this type of problem. More specifically, we show how the Value of Perfect Information (VPI) can be used to perform efficient decentralised exploration in both model-based and model-free BRL, and in the latter case, provide a closed form solution for VPI, correcting a decade old result by Dearden, Friedman and Russell. To evaluate these solutions, we present experimental results comparing their relative merits, and demonstrate empirically that both solutions outperform an existing multiagent learning method, representative of the state-of-the-art.

We analyse the problem of finding an allocation of resources in a multiagent system that is as fair as possible in terms of minimising inequality between the utility levels enjoyed by the individual agents. We use the well-known Atkinson index to measure inequality and we focus on the distributed approach to multiagent resource allocation, where new allocations emerge as the result of a sequence of local deals between groups of agents agreeing on an exchange of some of the items in their possession. Our results show that it is possible to design systems that provide theoretical guarantees for optimal outcomes that minimise inequality, but also that in practice there are significant computational hurdles to be overcome: finding an optimal allocation is computationally intractable---independently of the approach chosen---and large numbers of potentially highly complex deals may be required under the distributed approach. From a methodological point of view, while much work in multiagent resource allocation relies on combinatorial arguments, here we use insights from basic calculus.

This paper investigates a relatively new direction in Multiagent Reinforcement Learning. Most multiagent learning techniques focus on Nash equilibria as elements of both the learning algorithm and its evaluation criteria. In contrast, we propose a multiagent learning algorithm that is optimal in the sense of finding a best-response policy, rather than in reaching an equilibrium. We present the first learning algorithm that is provably optimal against restricted classes of non-stationary opponents. The algorithm infers an accurate model of the opponentýs non-stationary strategy, and simultaneously creates a best-response policy against that strategy. Our learning algorithm works within the very general framework of n-player, general-sum stochastic games, and learns both the game structure and its associated optimal policy.

The use of norms is widely accepted as an effective approach to control and regulate the behaviour of agents in multi-agent systems. Existing work on normative multi-agent systems has mainly focussed on how norms can influence the behaviour of agents by assuming that the agents' behaviours are perfectly monitored. In this paper we focus on monitoring mechanisms, propose different types of monitors, provide a logical analysis of monitors, study the relations between monitors and norms to be monitored, and finally explore computational aspects of norm monitoring.

A central problem in the study of autonomous cooperating systems is that of how to establish mechanisms for controlling the interactions between different parts (which are called agents) of the system. One way to integrate such mechanisms into a multi-agent system is to exploit the technique of cooperation or negotiation protocols. In a protocol we distinguish to essential layers: the communication layer specifying the possible flow of messages between different agents, and the decision layer, which controls the selection of a message (speech-act) that the agent sends in a specific situation. In this report we first give a short introduction of our agent model InteRRap which provides the basis for the modeling of the different scenarios considered in the AKA-Mod project at the DFKI. The techniques we will discuss in the following are located in the plan based component and in the cooperation component of this model. The domain of application is the MARS scenario (Modeling a Multi-Agent Scenario for Shipping Companies) which implements a group of shipping companies whose goal it is to deliver a set of dynamically given orders, satisfying a set of given time and/or cost constraints. The complexity of the orders may exceed the capacities of a single company. Therefore, cooperation between companies is required in order to achieve the goal in a satisfactory way. This domain is of considerable interest for studies with economical background as well as for research projects. We give a short summary of results from economical studies that are concerned with the real-world situation in Germany in the transportation domain. They show the need for the development of new techniques from the field of computer science to tackle the problems therein. Then, an overview on related research is presented. Two approaches are discussed in more detail: the first one being based on OR-techniques and a second one being based on the concept of partial intelligent agents attempting to integrate techniques from OR and DAI. Both approaches are concerned with the situation in a single company. However, our purpose to handle the case of distributed shipping companies requires additional mechanisms, e.g. to cope with the problems of task allocation and task decomposition in multi-agent systems. Mechanisms for distributed task decomposition and task allocation processes in multi-agent systems belong to the core of our studies. Therefore, we will first discuss techniques for these problems in a general setting and then describe their implementations in the MARS system. In this description, particular emphasis is placed on the cooperation within a shipping company. Here, one company agent has to allocate a set of orders its truck agents. The truck agents support the company agents by giving cost estimations based on their route planning facility. Thus, this procedure provides the basis for the decisions of the company agents and is discussed in very detail. Finally, we present results from a series of benchmark tests. The test sets have also been run with OR-implementations and thus, give us the opportunity to compare our implementation against these approaches.

Trust has been recognized as a key component of agent decision making in the context of multiagent systems (MASs). Though diverse trust models and mechanisms, influenced by various fields of study, have been proposed, implemented, and evaluated, we believe that the literature has ignored key aspects of pragmatic and holistic trust based reasoning. In particular, the focus of trust research has been on a posteriori evaluation of the trustworthiness of another agent and relatively few efforts have investigated the issue of establishment, engagement, and usage of trusted relationships. We envision that a holistic agent architecture will not use a trust module as a black-box for evaluating others but as a core component that will inform and shape interactions with other agents in the environment to best serve the decision-makers interests. Accordingly, we present a general and comprehensive trust management scheme (CTMS) that addresses key issues surrounding trust development, maintenance, and use. We present an operational definition of trust motivated by uncertainty management and utility optimization. We identify the various components required of a CTMS and their relationships and overview their use in the existing literature on trust in MAS. We welcome the MAS community to develop on the ideas presented here and build effective agent designs and implementations with fully-integrated CTMS cores.

Recommender systems are frequently used as part of online shops to help consumers browse through large product offerings by recommending those products which are the most relevant for them. Although consumers’ interactions with recommender systems have been subject to substantial research, it is still unclear what the effect on aggregated sales diversity is, i.e. whether this leads to predominance of fast-selling or niche products. It is also unclear, whether any potential effects would differ between specific recommender technologies. We created a realistic web-experiment to monitor consumer behavior while purchasing digital music tracks when different recommender technologies are present. To analyze potential changes in sales diversity we used the Gini coefficient as well as additional measures. We found that sales diversity increases for all recommender technologies, except for bestseller lists. Furthermore, the differences across recommender technologies are rather small. Our findings have significant implications for online retailers and for producers.

Recommender systems are an effective solution to the information overload problem, especially in the World Wide Web where we gather vast information from anonymous people around the world. Trust enhanced recommender system to be promising to overcome the cold-start and sparsity challenges of traditional recommender system as well as improving the accuracy of the recommendations. This arise a research focus about blending of the trust information and trust level prediction to the recommendation framework. From the past decade, numerous researches were done to adapt online social network trust (simply social trust) for many web applications, including e-commerce, P2P networks, multi-agent systems, recommendation systems, and service-oriented computing. Usually, online social trust prediction can be based on two mechanisms to acquire trust value: evaluating trustee on basis of truster / truster's neighbor trust experience information (local trust), otherwise evaluating trustee on the basis of the whole social network trust experience information as reputation (global trust). Here, we leverage social science theories to develop the trust models that enable the study of online social trust evolution. In this paper, we propose a matrix factorization based trust enhanced recommendation system which properly incorporates both local trust and global trust with diffusion of the social trust multi-aspects to improve the quality of recommendations for mitigating the data sparsity and the cold-start issues. Through experiments on the Epinions data set, we show that our model outperforms its standard trust enhanced counterparts with respect to accuracy on recommender systems.

Coalition structure generation (CSG) for multi-agent systems is a well-studied problem. A vast majority of the previous work and the state-of-the-art approaches to CSG assume a characteristic function form of the coalition values, where a coalition's value is independent of the other coalitions in the coalition structure. Recently, there has been interest in the more realistic partition function form of coalition values, where the value of a coalition is affected by how the other agents are partitioned, via externalities. We argue that in domains with externalities, a distributed/adaptive approach to CSG may be impractical, and that a centralized approach to CSG is more suitable. However, the most recent studies in this direction have focused on cases where all externalities are either always positive or always negative, and results on coalition structure generation in more general settings (in particular, mixed externalities) are lacking. In this paper we propose a framework based on agent-types that incorporates mixed externalities and demonstrate that it includes the previous settings as special cases. We also generalize some previous results in anytime CSG, showing that those results are again special cases. In particular, we extend the existing branch and bound algorithm to this new setting and show empirically that significant pruning can be achieved when searching for the optimal coalition structure. This extends the state-of-the-art in CSG for multi-agent systems.