Expert-Driven Validation of Rule-Based User Models in Personalization Applications

The last decades have seen an increasing interest in modeling collective animal behavior. Some studies try to reproduce as accurately as possible the collective dynamics and patterns observed in several animal groups with biologically plausible, individual behavioral rules. The objective is then essentially to demonstrate that the observed collective features may be the result of self-organizing processes involving quite simple individual behaviors. Other studies concentrate on the objective of establishing or enriching links between collective behavior researches and cognitive or physiological ones, which then requires that each individual rule be carefully validated. Here we discuss the methodological consequences of this additional requirement. Using the example of corpse clustering in ants, we first illustrate that it may be impossible to discriminate among alternative individual rules by considering only observational data collected at the group level. Six individual behavioral models are described: They are clearly distinct in terms of individual behaviors, they all reproduce satisfactorily the collective dynamics and distribution patterns observed in experiments, and we show theoretically that it is strictly impossible to discriminate two of these models even in the limit of an infinite amount of data whatever the accuracy level. A set of methodological steps are then listed and discussed as practical ways to partially overcome this problem. They involve complementary experimental protocols specifically designed to address the behavioral rules successively, conserving group-level data for the overall model validation. In this context, we highlight the importance of maintaining a sharp distinction between model enunciation, with explicit references to validated biological concepts, and formal translation of these concepts in terms of quantitative state variables and fittable functional dependences. Illustrative examples are provided of the benefits expected during the often long and difficult process of refining a behavioral model, designing adapted experimental protocols and inversing model parameters.

Personal agents gather information about users in a user profile. In this work, we propose a novel ontology-based user profile learning. Particularly, we aim to learn context-enriched user profiles using data mining techniques and ontologies. We are interested in knowing to what extent data mining techniques can be used for user profile generation, and how to utilize ontologies for user profile improvement. The objective is to semantically enrich a user profile with contextual information by using association rules, Bayesian networks and ontologies in order to improve agent performance. At runtime, we learn which the relevant contexts to the user are based on the user's behavior observation. Then, we represent the relevant contexts learnt as ontology segments. The encouraging experimental results show the usefulness of including semantics into a user profile as well as the advantages of integrating agents and data mining using ontologies.

Mobile device can be used as a medium to send and receive the mobile internet content. However, there are several limitations using mobile internet. Content personalisation has been viewed as an important area when using mobile internet. In order for personalisation to be successful, understanding the user is important. In this paper, we explore the implementation of the user profile at client-side, which may be used whenever user connect to the mobile content provider. The client-side user profile can help to free the provider in performing analysis by using data mining technique at the mobile device. This research investigates the conceptual idea of using clustering and classification of user profile at the client-site mobile. In this paper, we applied K-means and compared several other classification algorithms like TwoStep, Kohenen and Anomaly to determine the boundaries of the important factors using information ranking separation.

To reduce the threshold of EV adoption, the electric vehicle supply equipment (EVSE) infrastructure needs to fit the EV systems to its users to satisfy their expectations. There needs to be a reflection of user behavior variability in the EVSE infrastructure to improve its functionality. Individual users’ charging actions, over time, construct a charging behavioral pattern that distinguishes the users from one another. This research analyses the EV users’ unique behavioral charging patterns. Therefore, profiles of users’ charging actions are clustered to investigate the unique behavioral patterns. An unsupervised clustering algorithm (i.e., K-means) is implemented through three distance metrics (i.e., Hamming, Wasserstein, and Manhattan) to classify the Rhode Island public charging stations’ frequent users. Frequent users of these stations were established using the Pareto Principle, in which 608 users (20% of users) contributed to 89% of charging events. The results indicated five clusters through the Wasserstein distance metric, which proposed different EV charging behavior patterns. The five achieved clusters represent: 39% of anxious or opportunistic users (1); 27% of users with consistent charging regardless of their EVs state of charge (2); while 21% of users having sporadic charging behavior revealing no pattern (3); and then approximately 5% of users with procrastination tendencies with only charging when practically out of charge (4) and 8% of users who experience that rush toward running out of charge (procrastinators) but exhibit some opportunistic early charging (5). Knowing these users’ unique behaviors helps both public and private stakeholders target EV market growth through user-centric EVSE location placement.

Nowadays health concerns are effectively becoming ubiquitous. Most people have the need to effectively control their nutritional consumptions, mostly due to health issues. Personal computational devices may assist this control with a solution that allows an efficient management of each individual nutritional profile. In this work we propose a mobile service architecture that allows users to manage their nutritional information, using a profile-based system and build shopping lists based on the user's profiles. This application may contribute to the improvement of the lifestyle of the population through the recommendation of food and drinks that fit their profile of restrictions and/or nutritional options (for instance, due to hypertension or obesity, among others). The person's profile can be accessed and configured on a mobile device. A set of predefined templates provides the initial rules that may be customized to represent specific individual nutrition rules. The rules defined in the profile can later be used to filter the food presented to each user. The paper includes preliminary usability results from experiments using real data to validate the approach. These results suggest it could be used in real scenarios, although it may require more than simply filtering results.

In e-commerce it is often crucial to provide customers a large choice of relevant offers. Users, however, seldom provides complete and comprehensive descriptions of their desires, therefore user interfaces are needed that can generate automatically expanded queries to the product database and proactively enrich the ongoing dialogue with recommendations of suitable products. Automatic query expansion is mostly based on thesaurus and/or user profiles. In e-commerce applications, specific thesauri reflecting the webstore's product categories are desirable. This work describes a method for the automatic construction of a thesaurus based on existing categories of documents. A clustering algorithm, the "Layer-Seeds method'', is introduced, which facilitates the automatic generation of thesaurus reflecting the specific vocabulary occurring in a given collection of documents. The clustering works on terms extracted from the documents in a certain category and organizes them in a tree-like hierarchical structure--a thesaurus. The thesaurus is then employed for automatic query expansion in an e-commerce application in order to obtain better results for product searching. Experiments yield evidence that a significant increase of user satisfaction is achieved.

Cloud Computing is growing exponentially across organizations in various domains and it has a vast impact on the way software gets developed and tested. The web based applications these days have configuration settings different from deployment requirements. The main focus of Cloud Computing is to deliver reliable, secured, fault-tolerant and elastic infrastructures for hosting an E-Commerce application. Scheduling policies and allocation policies for resources which affect the performance and utilization of cloud infrastructure (i.e. hardware, software services) for various E-Commerce application under varying load and system size is highly challenging problem to deal with. Performance analysis and optimal resource management policies allows cloud service providers to improve their Quality of Service (QOS). This work focuses on the process of selecting the best resource estimation policy for a given workload from various policies such as, Maximum Log-Likelihood, Maximum Product of Spacing Estimator, and Probability Weighted Movements. Detailed experimentation has been carried out for using the Amazon Web Service(AWS) public cloud, mimicking it on the cloudsim simulator. The modeling of workload is quiet challenging due to the unavailability of trace logs for analysis. Hence, in this paper, workload is generated based on application model for E-commerce application considering varying user behavior relating to different user profiles. The amount of resources consumed is closely monitored and a resource usage model has been developed and validated to choose the best estimation policy.

Releasing social network data could seriously breach user privacy. User profile and friendship relations are inherently private. Unfortunately, sensitive information may be predicted out of released data through data mining techniques. Therefore, sanitizing network data prior to release is necessary. In this paper, we explore how to launch an inference attack exploiting social networks with a mixture of non-sensitive attributes and social relationships. We map this issue to a collective classification problem and propose a collective inference model. In our model, an attacker utilizes user profile and social relationships in a collective manner to predict sensitive information of related victims in a released social network dataset. To protect against such attacks, we propose a data sanitization method collectively manipulating user profile and friendship relations. Besides sanitizing friendship relations, the proposed method can take advantages of various data-manipulating methods. We show that we can easily reduce adversary's prediction accuracy on sensitive information, while resulting in less accuracy decrease on non-sensitive information towards three social network datasets. This is the first work to employ collective methods involving various data-manipulating methods and social relationships to protect against inference attacks in social networks.

In order to adapt functionality to their individual users, systems need information about these users. The Social Web provides opportunities to gather user data from outside the system itself. Aggregated user data may be useful to address cold-start problems as well as sparse user profiles, but this depends on the nature of individual user profiles distributed on the Social Web. For example, does it make sense to re-use Flickr profiles to recommend bookmarks in Delicious? In this article, we study distributed form-based and tag-based user profiles, based on a large dataset aggregated from the Social Web. We analyze the completeness, consistency and replication of form-based profiles, which users explicitly create by filling out forms at Social Web systems such as Twitter, Facebook and LinkedIn. We also investigate tag-based profiles, which result from social tagging activities in systems such as Flickr, Delicious and StumbleUpon: to what extent do tag-based profiles overlap between different systems, what are the benefits of aggregating tag-based profiles. Based on these insights, we developed and evaluated the performance of several cross-system user modeling strategies in the context of recommender systems. The evaluation results show that the proposed methods solve the cold-start problem and improve recommendation quality significantly, even beyond the cold-start.

The recent improvements in smartphones nowadays offer a widespread application of sensor-based services. Each mobile phone is equipped with several sensors like a GPS module, a gyroscope, or a high-resolution camera. As a result of this sensor integration, a whole new way of usage is opened up for the end-user, like a location-based search or people-centric sensing. The main drawback related to a smartphone is an overall high energy consumption, combined with a limited energy capacity. Due to this fact, a continuous and fine grained sensing of the user's context is not possible, as it utilizes at least one acceleration sensor. Furthermore, the captured data is transmitted via a (mobile) communication infrastructure to post the context on the Internet. Both drain the battery very quickly. For that reason, an efficient energy-constrained distribution is required to minimize the update occurrence of a producer, while simultaneously maximizing the accuracy of a consumer. The primary issues to be addressed include a modeling of user behavior as well as a determination of optimal points in time for an update. Therefore, a probabilistic approach is used to forecast the user's context pattern. The prediction is based upon a Markov chain and enables the extraction of meaningful information. The proper times for an update are determined with the help of a constrained optimization problem. Different methods from mathematical optimization are applied like linear and nonlinear programming or a constrained Markov decision process, which obtain an update policy. For a better comparison of the weaknesses and strengths related to the developed methods, dynamic programming is used to achieve the optimal points in time for an update. The evaluation upon a real trace shows that an accuracy gain of more than 30% is achieved by sending the equal amount of messages.

PurposeThis paper aims to propose a solution for recommending digital library services based on data mining techniques (clustering and predictive classification).Design/methodology/approachData mining techniques are used to recommend digital library services based on the user's profile and search history. First, similar users were clustered together, based on their profiles and search behavior. Then predictive classification for recommending appropriate services to them was used. It has been shown that users in the same cluster have a high probability of accepting similar services or their patterns.FindingsThe results indicate that k‐means clustering and Naive Bayes classification may be used to improve the accuracy of service recommendation. The overall accuracy is satisfying, while average accuracy depends on the specific service. The results were better for frequently occurring services.Research limitations/implicationsDatasets were used from the KOBSON digital library. Only clustering and predictive classification was applied. If the correlation between the service and the institution were higher, it would have better accuracy.Originality/valueThe paper applied different and efficient data mining techniques for clustering digital library users based on their profiles and their search behavior, i.e. users' interaction with library services, and obtain user patterns with respect to the library services they use. A digital library may apply this approach to offer appropriate services to new users more easily. The recommendations will be based on library items that similar users have already found useful.

Recommender systems combine research from user profiling, information filtering and artificial intelligence to provide users with more intelligent information access. They have proven to be useful in a range of Internet and e-commerce applications. Recent research has shown that a content-based (or case-based) perspective on collaborative filtering for recommendation can provide significant benefits in decision support accuracy over traditional collaborative techniques, particularly as dataset sparsity increases. These benefits derive both from the use of more sophisticated case-based similarity metrics and from the proactive maintenance of item similarity knowledge using data mining. This article presents a natural next step in this ongoing research to improve the quality of recommender systems by validating these findings in the context of more complex models of collaborative filtering, as well as by demonstrating that such techniques also preserve recommendation diversity, one of the key issues affecting traditional recommender systems.

Research into session-based recommendation systems (SBSR) has attracted a lot of attention, but each study focuses on a specific class of methods. This work examines and evaluates a large range of methods, from simpler statistical co-occurrence methods to embeddings and SotA deep learning methods. This paper analyzes theoretical and practical issues in developing and evaluating methods for SBSR in e-commerce applications, where user profiles and purchase data do not exist. The major tasks of SBRS are reviewed and studied, namely: prediction of next-item, next-basket and purchase intent. For physical retail shopping where no information about the current session exists, we treat the previous baskets purchased by the user as previous sessions drawn from a loyalty system. Mobile application scenarios such as push notifications and calling tune recommendations are also presented. Recommender models using graphs, embeddings and deep learning methods are studied and evaluated in all SBRS tasks using different datasets. Our work contributes a number of very interesting findings. Among all tested models, LSTMs consistently outperform other methods of SBRS in all tasks. They can be applied directly because they do not need significant fine-tuning. Additionally, they naturally model the dynamic browsing that happens in e-commerce web applications. On the other hand, another important finding of our work is that graph-based methods can be a good compromise between effectiveness and efficiency. Another important conclusion is that a “temporal locality principle” holds, implying that more recent behavior is better suited for prediction. In order to evaluate these systems further in realistic environments, several session-based recommender methods were integrated into an e-shop and an A/B testing method was applied. The results of this A/B testing are in line with the experimental results, which represents another important contribution of this paper. Finally, important parameters such as efficiency, application of business rules, re-ranking issues, and the utilization of hybrid methods are also considered and tested, providing comprehensive useful insights into SBRS and facilitating the transferability of this research work to other domains and recommendation scenarios.

Web based commercial recommender systems (RS) can help users to make decisions about which product to purchase from the vast amount of products available on the Internet. Currently, many commercial recommender systems are developed for recommending frequently purchased products where a large amount of explicit ratings or purchase history data is available to predict user preferences. However, for products that are infrequently purchased by users, it is difficult to collect such data and, thus, user profiling becomes a major challenge for recommending these kinds of products. This paper proposes a recommendation approach for infrequently purchased products based on user opinions and navigation data. User opinion data, which is collected from product review data, is used to generate product profiles and user navigation data is used to generate user profiles, both of which are used for recommending products that best satisfy the users' needs. Experiments conducted on real e-commerce data show that the proposed approach, named, Adaptive Collaborative Filtering (ACF), which utilizes user and product profiles, outperforms the Query Expansion (QE) approach that only utilizes product profiles to recommend products. The ACF also performs better than the Basic Search (BS) approach, which is widely applied by the current e-commerce applications.

Extreme outside temperatures resulting from heat waves, winter storms, and similar weather-related events trigger the Heating Ventilation and Air Conditioning (HVAC) systems, resulting in challenging, and potentially catastrophic, peak loads. As a consequence, such extreme outside temperatures put a strain on power grids and may thus lead to blackouts. To avoid the financial and personal repercussions of peak loads, demand response and power conservation represent promising solutions. Despite numerous efforts, it has been shown that the current state-of-the-art fails to consider (1) the complexity of human behavior when interacting with power conservation systems and (2) realistic home-level power dynamics. As a consequence, this leads to approaches that are (1) ineffective due to poor long-term user engagement and (2) too abstract to be used in real-world settings. In this article, we propose an auction theory-based power conservation framework for HVAC designed to address such individual human component through a three-fold approach: personalized preferences of power conservation, models of realistic user behavior , and realistic home-level power dynamics . In our framework, the System Operator sends Load Serving Entities (LSEs) the required power saving to tackle peak loads at the residential distribution feeder. Each LSE then prompts its users to provide bids , i.e., personalized preferences of thermostat temperature adjustments, along with corresponding financial compensations. We employ models of realistic user behavior by means of online surveys to gather user bids and evaluate user interaction with such system. Realistic home-level power dynamics are implemented by our machine learning-based Power Saving Predictions (PSP) algorithm, calculating the individual power savings in each user’s home resulting from such bids. A machine learning-based PSPs algorithm is executed by the users’ Smart Energy Management System (SEMS). PSP translates temperature adjustments into the corresponding power savings. Then, the SEMS sends bids back to the LSE, which selects the auction winners through an optimization problem called POwer Conservation Optimization (POCO). We prove that POCO is NP-hard, and thus provide two approaches to solve this problem. One approach is an optimal pseudo-polynomial algorithm called DYnamic programming Power Saving (DYPS), while the second is a heuristic polynomial time algorithm called Greedy Ranking AllocatioN (GRAN). EnergyPlus, the high-fidelity and gold-standard energy simulator funded by the U.S. Department of Energy, was used to validate our experiments, as well as to collect data to train PSP. We further evaluate the results of the auctions across several scenarios, showing that, as expected, DYPS finds the optimal solution, while GRAN outperforms recent state-of-the-art approaches.