A new way to train a virtual assistant with unsupervised learning is presented in this thesis. Rather than integrating with a particular set of programs and interfaces, this new approach involves shallow integration between the virtual assistant and computer through machine vision. In effect the assistant interprets the computer screen in order to produce helpful recommendations to assist the computer user. In developing this new approach, called AVRA, the following methods are described: an unsupervised learning algorithm which enables the system to watch and learn from user behavior, a method for fast filtering of the text displayed on the computer screen, a deep learning classifier used to recognize key onscreen text in the presence of OCR translation errors, and a recommendation filtering algorithm to triage the many possible action recommendations. AVRA is compared to a similar commercial state-of-the-art system, to highlight how this work adds to the state of the art. AVRA is a deep learning image processing and recommender system that can col- laborate with the computer user to accomplish various tasks. This document presents a comprehensive overview of the development and possible applications of this novel vir- tual assistant technology. It detects onscreen tasks based upon the context it perceives by analyzing successive computer screen images with neural networks. AVRA is a rec- ommender system, as it assists the user by producing action recommendations regarding onscreen tasks. In order to simplify the interaction between the user and AVRA, the system was designed to only produce action recommendations that can be accepted with a single mouse click. These action recommendations are produced without integration into each individual application executing on the computer. Furthermore, the action recommendations are personalized to the user’s interests utilizing a history of the user’s interaction.

In typical restaurant recommendations, knowledge-based methods are used most often and do not take advantage of personal historical data. In this thesis, we are going to make some improvements to the Chicago Entrée restaurant recommender system. We will exploit the historical data and propose a weighted similarity approach to combine heuristic similarity with tag similarity between restaurants. Also, we show an improved way to mine the semantics of user behaviors using heuristic metric. These proposed approaches are evaluated by the comparison of three different pairwise approaches to learning to rank (LTR) in matrix factorization and five classic recommendation algorithms. The result shows that the combinatorial similarity outperforms the heuristic similarity on the precision, recall, F-score, and mean reciprocal rank.

In general, recommender systems attempt to estimate user preference based on historical data. A context-aware recommender system attempts to generate better recommendations using contextual information. However, generating recommendations for specific contexts has been challenging because of the difficulties in using contextual information to enhance the capabilities of recommender systems. Several methods have been used to incorporate contextual information into traditional recommendation algorithms. These methods focus on incorporating contextual information to improve general recommendations for users rather than identifying the different context applicable to the user and providing recommendations geared towards those specific contexts. In this thesis, we explore different context-aware recommendation techniques and present our context-aware coupled matrix factorization methods that use matrix factorization for estimating user preference and features in a specific contextual condition. We develop two methods: the first method attaches user preference across multiple contextual conditions, making the assumption that user preference remains the same, but the suitability of items differs across different contextual conditions; i.e., an item might not be suitable for certain conditions. The second method assumes that item suitability remains the same across different contextual conditions but user preference changes. We perform a number of experiments on the last.fm dataset to evaluate our methods. We also compared our work to other context-aware recommendation approaches. Our results show that grouping ratings by context and jointly factorizing with common factors improves prediction accuracy.

abstract: Currently, recommender systems are used extensively to find the right audience with the "right" content over various platforms. Recommendations generated by these systems aim to offer relevant items to users. Different approaches have been suggested to solve this problem mainly by using the rating history of the user or by identifying the preferences of similar users. Most of the existing recommendation systems are formulated in an identical fashion, where a model is trained to capture the underlying preferences of users over different kinds of items. Once it is deployed, the model suggests personalized recommendations precisely, and it is assumed that the preferences of users are perfectly reflected by the historical data. However, such user data might be limited in practice, and the characteristics of users may constantly evolve during their intensive interaction between recommendation systems. Moreover, most of these recommender systems suffer from the cold-start problems where insufficient data for new users or products results in reduced overall recommendation output. In the current study, we have built a recommender system to recommend movies to users. Biclustering algorithm is used to cluster the users and movies simultaneously at the beginning to generate explainable recommendations, and these biclusters are used to form a gridworld where Q-Learning is used to learn the policy to traverse through the grid. The reward function uses the Jaccard Index, which is a measure of common users between two biclusters. Demographic details of new users are used to generate recommendations that solve the cold-start problem too. Lastly, the implemented algorithm is examined with a real-world dataset against the widely used recommendation algorithm and the performance for the cold-start cases. / Dissertation/Thesis / Masters Thesis Computer Science 2020

To combat fake news, researchers mostly focused on detecting fake news and journalists built and maintained fact-checking sites (e.g., Snopes.com and Politifact.com). However, fake news dissemination has been greatly promoted by social media sites, and these fact-checking sites have not been fully utilized. To overcome these problems and complement existing methods against fake news, in this thesis, we propose a deep-learning based fact-checking URL recommender system to mitigate impact of fake news in social media sites such as Twitter and Facebook. In particular, our proposed framework consists of a multi-relational attentive module and a heterogeneous graph attention network to learn complex/semantic relationship between user-URL pairs, user-user pairs, and URL-URL pairs. Extensive experiments on a real-world dataset show that our proposed framework outperforms seven state-of-the-art recommendation models, achieving at least 3~5.3% improvement.

The internet is larger than ever and so is the amount of information on the internet.The average user on the internet has next to endless possibilities and choices whichcan cause information overload. Companies have therefore developed systems toguide their users to find the right product or object in the form of recommendersystems. Recommender systems are tools created to filter data and find patternsto recommend relevant information for specific customers with the help of differentalgorithms. MarketHype is a company that aggregates large amounts of data aboutevent organizers, their events, their visitors, and related transactions. They want inthe near future to be able to manage and offer event organizers recommended targetgroups for their events using a recommender system.This study tries to find a solution on how to model event data in a graph databaseto support relevant recommendations for event organizers. The method used to answer the question was an empirical research method. The goal was to create aprototype of a recommender system with help of event data. The main focus was tomodel a graph database in the software Neo4j that can be used for finding recommendations with different Cypher queries. A literature study was later conducted tofind what advantages and disadvantages a graph database could have on event data.This information could then answer how further development of the system couldwork.The result was a system that was implemented with the help of data from fourdifferent CSV files. The data provided were information about contacts, persons,orders, and events. This information was used to create the nodes and relationships.A total of 4.4 million nodes were created and around 5 million relationships betweenthose nodes. Collaborative and content-based filtering was the main recommendationtechnique used in order to find the best-suitable recommendations. This was donewith different queries in Cypher.The main conclusion is that a graph database in Neo4j is a good method in orderto implement a recommender system with event data. The result shows that thecollaborative filtering approach is a major factor in the system’s success in findingrelevant information. The approach of letting other contacts decide what the originalcontract wants is proven to work well with event data. The result also states thatthe recommendation is more of an indication because it returns what supposedlywould be the preferences for a contact. A solution for a better recommender systemwas found which includes another layer to the content-based filtering in the form ofcategorized events.

An individual’s ability to locate the information they desire grows more slowly than the rate at which new information becomes available. Customers are constantly confronted with situations in which they have many options to choose from and need assistance exploring or narrowing down the possibilities. Recommender systems are one tool to help bridge this gap. There are various mechanisms being employed to create recommender systems, but the most common systems fall into two main classes: content-based and collaborative filtering systems. Content-based recommender systems match the textual information of a particular product with the textual information representing the interests of a customer. Collaborative filtering systems use patterns in customer ratings to make recommendations. Both types of recommender systems require significant data resources in the form of a customer’s ratings and product features; hence they are not able to generate high quality recommendations. Hybrid mechanisms have been used by researchers to improve the performance of recommender systems where one can integrate more than one mechanism to overcome the drawbacks of an individual system. The hybrid approach proposed in this thesis is the integration of content and context-based with collaborative filtering, since these are the most successful and widely used mechanisms. This proposed approach will look into the integration of content and context data with rating data using a different mechanism that mainly focuses on boosting a customer’s trust in the recommender system. Researchers have been trying to improve system performance using hybrid approaches, but research is lacking on providing justifications for recommended products. Hence, the proposed approach will mainly focus on providing justifications for recommended products as this plays a crucial role in obtaining the satisfaction and trust of customers. A product’s features and a customer’s context attributes are used to provide justifications. In addition to this, the presentation mechanism needs to be very effective as it has been observed that customers trust more in a system when there are explanations on how the recommended products have been computed and presented. Finally, this proposed recommender system will allow the customer to interact with it in various ways to provide feedback on the recommendations and justifications. Overall, this integration will be very useful in achieving a stronger correlation between the customers and products. Experimental results clearly showed that the majority of the participants prefer to have recommendations with their justifications and they received valuable recommendations on which they could trust.

Recommender systems are apparent in our lives through multiple different ways, such asrecommending what items to purchase when online shopping, recommending movies towatch and recommending restaurants in your area. This thesis aims to apply the sametechniques of recommender systems on a new area, namely stock recommendations basedon your current portfolio. The data used was collected from a social media platform forinvestments, Shareville, and contained multiple users portfolios. The implicit data wasthen used to train matrix factorization models, and the state-of-the-art LightGCN model.Experiments regarding different data splits was also conducted. Results indicate that rec-ommender systems techniques can be applied successfully to generate stock recommen-dations. Also, that the relative performance of the models on this dataset are in line withprevious research. LightGCN greatly outperforms matrix factorization models on this pro-posed dataset. The results also show that different data splits also greatly impact the re-sults, which is discussed in further detail in this thesis.

In the study, two methods for recommending application software were implemented and evaluated based on their ability to recommend alternative applications with related functionality to the one that a user is currently browsing. One method was based on Term Frequency–Inverse Document Frequency (TF-IDF) and the other was based on Latent Semantic Indexing (LSI). The dataset used was a set of 2501 articles from Wikipedia, each describing a distinct application. Two experiments were performed to evaluate the methods. The first experiment consisted of measuring to what extent the recommendations for an application belong to the same software category, and the second was a set of structured interviews in which recommendations for a subset of the applications in the dataset were evaluated more in-depth. The results from the two experiments showed only a small difference between the methods, with a slight advantage to LSI for smaller sets of recommendations retrieved, and an advantage for TF-IDF for larger sets of recommendations retrieved. The interviews indicated that the recommendations from when LSI was used to a higher extent had a similar functionality as the evaluated applications. The recommendations from when TF-IDF was used had a higher fraction of applications with functionality that complemented or enhanced the functionality of the evaluated applications. / I studien implementerades och utvärderades två alternativa implementationer av ett rekommendationssystem för applikationsprogramvara. Implementationerna utvärderades baserat på deras förmåga att föreslå alternativa applikationer med relaterad funktionalitet till den applikation som användaren av ett system besöker eller visar. Den ena implementationen baserades på Term Frequency-Inverse Document Frequency (TF-IDF) och den andra på Latent Semantic Indexing (LSI). Det data som användes i studien bestod av 2501 artiklar från engelska Wikipedia, där varje artikel bestod av en beskrivning av en applikation. Två experiment utfördes för att utvärdera de båda metoderna. Det första experimentet bestod av att mäta till vilken grad de rekommenderade applikationerna tillhörde samma mjukvarukategori som den applikation de rekommenderats som alternativ till. Det andra experimentet bestod av ett antal strukturerade intervjuer, där rekommendationerna för en delmängd av applikationerna utvärderades mer djupgående. Resultaten från experimenten visade endast en liten skillnad mellan de båda metoderna, med en liten fördel till LSI när färre rekommendationer hämtades, och en liten fördel för TF-IDF när fler rekommendationer hämtades. Intervjuerna visade att rekommendationerna från den LSI-baserade implementationen till en högre grad hade liknande funktionalitet som de utvärderade applikationerna, och att rekommendationerna från när TF-IDF användes till en högre grad hade funktionalitet som kompletterade eller förbättrade de utvärderade applikationerna.

Recommender systems apply machine learning methods to solve the task of providing appropriate suggestions to users in both static and dynamic environments. An example of this is a movie service like Netflix that recommends movies to its users. Although many algorithms have been proposed, making predictions for users with few ratings remains a challenge in recommender systems. In this study the performance of the algorithm k-NN, both user- and item-based, was empirically evaluated. This was done using the MovieLens 1M and 100K datasets in scenarios where the users have between 1 and 9 ratings, simulating cold-start scenarios of various degree. The results were then compared with the accuracy of the algorithm in a simulated normal case, to see how the cold-start affected the two algorithms, and which one of them that handled it best. In summary, this report shows that user-based k-NN performs better in relation to item-based k-NN for new users having few rated items. Overall the accuracy improved as the number of ratings increased for the new users for both user- and item-based k-NN.