The purpose of this report is to make a comparison of how fuzzy logic rules affect the traffic flow in an intersection in the inner city of Stockholm. For the comparison, the report uses a custom written simulation to answer the question and the focus is primarily on measuring the difference in waiting time for the cars in the queues and the difference in the length of the queues. For this we have built our own fuzzy logic rules, specially adapted for the structure of the selected intersection. They were then implemented based on the current precalculated green times for the intersection. The data used for the simulation is collected by the Stockholm stad during November 2015. Based on the work it can be concluded that the fuzzy logic rules makes the intersection more efficient, especially in terms of the average queue length which during almost the entire simulation is shorter than without the use of fuzzy logic rules. With respect to the average waiting time the result is not quite as clear, but it is possible to justify that it becomes more effective using the fuzzy logic rules. From these two statements one can fairly easily observe that using fuzzy logic can be used to streamline the flow of traffic at the intersection, and also to streamline traffic with respect to delays.

Denna studie undersökte hur en tillämpning av myrarten Atta Colombicas trafikregler påverkade trafikflödet på en dubbelriktad väg med två körfält där en del av ett körfält var avstängt. Detta gjordes genom att utveckla ett simulationsprogram, en algoritm baserad på Atta Colombicas trafikregler samt en algoritm baserad på svenska trafikregler. Algoritmerna jämfördes med avseende på trafikflöde och fordons genomsnittliga resetid. Simuleringen samlade kvantitativ data om trafikflödet och genomsnittliga resetider från båda algoritmer. Resultaten visade att algoritmen baserad på svenska trafikregler presterade bättre i alla testfall. Slutsatsen att Atta Colombicas trafikregler inte ökar flödet jämfört med svenska trafikregler drogs.

An area where algorithms are used in order to optimize prediction of what users prefer in di↵erent market areas. By modifying algorithms that has already been made, one can improve the accuracy of the predictions made. The purpose of this study was to find out if existing behavioral prediction algorithms can be improved by taking certain parameters into consideration. This was achieved by conducting a survey where users could vote on what parameters they take into consideration when chosing a movie to watch. Once having the important parameters, a modification was made to an existing algorithm so that a comparison in di↵erences was possible. The results were clear and indicated that the modification was very data-reliant and therefore not optimal for every situation.

Content providers require an efficient and accurate way of retrieving relevant content with minimal human aid. News retrieval, for instance, often requires human intervention to recognize which text documents are news articles and which are not. The differences between a factual news article and an opinionated blog piece may be subtle, yet are critical for providing informative and relevant content to users. This thesis explores the problem of format classification: the task of classifying text documents based on the format in which they are written, such as a news article, blog entry or forum text. More explicitly, the goal of the thesis is to examine how well state-of-the-art supervised text classifica- tion techniques work for format classification. We select a number of classifiers that have been shown to perform well in other text classification tasks and evaluate their perfor- mance in this unexplored task. Experimental evaluation, performed on a novel dataset created from multiple existing datasets, explores both binary and multi-class classification in a bag-of-words feature space. Based on our experimental results, we have found that state-of-the-art supervised text classification techniques perform acceptably well at format classification. Furthermore, we propose a Gradient Boost model as a candidate classifier for the task of format clas- sification, and provide a discussion of future work. / Företag som tillhandahåller innehållshanteringstjänster be- höver effektiva och precisa metoder för att med minsta möjliga mänskliga arbetskraft utvinna relevant innehåll ur stora mängder data. Ett exempel på detta är tjänster för insamlande av nyheter, där nyheter skall utvinnas från olika källor. Som en del av den processen krävs att de kan avgöra om en text är en nyhetsartikel eller någon annan form av text. Skillnaden mellan en nyhetsartikel och en text skriven för en blogg kan vara subtil, men är avgörande för dessa företag. Denna rapport utforskar formatklassifi- cering: uppgiften att klassificera texter baserat på vilket format de är skrivna för. Exempel på format är: nyhet- sartikel, bloggtext eller forumtext. Mer specifikt tar den sig an uppgiften att undersöka hur väl de metoder som idag används i den väl utforskade uppgiften att klassificera texter baserat på ämne fungerar applicerade på formatk- lassificering. Det utforskas med experimentell evaluering på ett nytt dataset som konstruerats genom att kombin- era ett flertal existerande dataset. Detta görs både som en binär- och multiklassificeringsuppgift i en bag-of-word vektorrymd. Ett antal ämnesklassificeringsmetoder väljs baserat på resultat från tidigare forskning, och hur dessa presterar på formatklassificering undersöks. Vi drar slut- satsen att våra resultat visar att de textklassificeringsme- toder vi testat fungerar acceptabelt väl på formatklassifi- cering. Vi föreslår vidare gradient-boost eller multinomial naive bayes för att lösa uppgiften, beroende på om fokus ligger på kvaliteten av klassificeringen eller prestanda. Slut- ligen diskuteras resultaten , de sätts i relation till de begrän- sningar som förelegat och förslag till framtida forskning ges.

Sociala medier har alltid varit en plats där personer kan diskutera fritt om sina åsikter och dela nyheter med många. Lätt spridning av nyheter från alla hörn i världen kan komma vara användbart för att ha möjlighet att få opartiska nyheter. Även om det finns klara fördelar med exempelvis Twitter så kan det vara problematiskt med falska och uppgjorda nyheter. Ryktesspridning eller uppgjorda nyheter förekommer i stor utsträckning fortfarande, här testas metod(er) för att upptäcka vilka som kan tänkas sprida desinformation, så kallade ’trolls’. För att kunna upptäcka trolls i Twitter undersöks några attributer som tidigare använts för att detektera spammare. Problem uppstår då det inte finns några fastställda troll att jämföra med. Datasamling från Twitter utfördes och analyserats, genom att klustra data med viktiga attribut som skulle indikera på förekomst av trolls som exempelvis antalet tweets varje dag eller hur ofta det retweetas. Men klustringen ger inte 100% indikation på att det finns trolls utan att det skulle kunna öka sannolikheten att hitta trolls i ett kluster där en typisk twitterkonto har hög andel retweets eller följare genom vänner. Slutsatsen blev att metoden kan inte hitta trolls på egen hand men kan hjälpa till att isolera användare med specifika beteenden som kan vara mer eller mindre troll-lika. / Social media has always been a place where people can discuss freely about their opinions and share news with many. Easy dissemination of news from all corners of the world may be useful to be able to get unbiased news. But although there are clear advantages, for example, with Twitter it can be problematic with false and rigged news. Rumor or news-fixing exists and to a large extent still is, this report tests method(s) to detect those that might spread disinformation, so-called ’trolls’. In order to detect a troll in Twitter some attributes previously been used to detect spammers, are being examined. Problems arise as a lack of ground truth facts to compare and back the results with. Data Collection from Twitter were performed and analyzed by clustering the data with the key attributes that would indicate the presence of trolls such as the number of tweets each day or how often it retweets. However, clustering does not give 100% indication that there are trolls but it could increase the probability of finding troll in a cluster where a typical Twitter account has high proportion retweets or followers through friends. The conclusion was that the method can not find trolls by its own. What it can do is isolate specific behaviors that can be more or less troll-like.

In modern society the amount of information processed by computers is increasing everyday. Computer translation has the potential to speed up communication between humans as well as human-computer interactions. For Statistical Machine Translation word alignment is key. How large does a corpus need to be to align a natural language sentence with a simple unambiguous language? We investigate this matter by running a simple algorithm and comparing it to the results we get from an industry equivalent. The results show that the size of the corpus needs to be larger for the simplified model when there is a greater number of words per sentence. The IBM Model 4 conversely shows that the more words per sentence decrease the necessary size of the corpus to make better predictions.Thus we can conclude that corpus size is dependant on the number of terms in each sentence for both models. / I vårat moderna samhälle bearbetas mer information för varje dag. Datoriserad översättning har potentialen att öka hastigheten utav kommunikationen mellan människor emellan samt människa-datorinteraktion. För Statistical Machine Translation så är word alignment en stor del. Hur stor måste en korpus vara för att man med stor sannolikhet lyckats att korrekt översätta meningar från ett naturligt språk med ett simpelt entydigtspråk?Vi testar detta genom att jämföra en simpel algorithm med en algoritm som används inom industrin. I resultaten ser vi att ju mer ord som finns i meningen som ska översättas, ju större måste korpusen vara. Med IBM Model 4 ser vi att resultaten blir bättre med ju fler ord per mening och därför kan korpusstorleken minskas. Vår slutsats är att korpus storleken beror på mängden aritmetiska termer för båda modellerna.

In recent years, reading text from natural images has gained renewed research attention. One of the main reasons for this is the rapid growth of camera-based applications on smart phones and other portable devices. With the increasing availability of high performance, low-priced, image-capturing devices, the application of scene text recognition is rapidly expanding and becoming increasingly popular. Despite many efforts, character recognition in natural images, is still considered a challenging and unresolved problem. The difficulties stem from the fact that natural images suffer from a wide variety of obstacles such as complex backgrounds, font variation, uneven illumination, resolution problems, occlusions, perspective effects, just to mention a few. This paper aims to test the accuracy of OCR in character recognition of natural images as well as testing the possible improvement in accuracy after implementing three different segmentation methods.The results showed that the accuracy of OCR was very poor and no improvments in accuracy were found after implementing the chosen segmentation methods.

Predicting the product a customer would like to buy is an increasingly important field of study and there are several different recommender system models that are used to make recommendations for users. Deep learning has shown effective results in a variety of predictive tasks but there haven’t been much research concerning its usage in recommender systems. This thesis studies the effectiveness of using a long short term memory implementation (LSTM) of a recurrent neural network (RNN) as a recommender system by comparing it to one of the most common recommender system implementations, the matrix factorization method. A radio playlist dataset is used to train both the LSTM and the matrix factorization models with the intent of generating accurate predictions. We were unable to create a LSTM model with good performance and due to that we are unable to make any significant conclusions regarding whether or not LSTM networks outperform matrix factorization models.

Developing a Natural Language Interface that can understand everything is a very challenging task due to the varied and ambiguous nature of natural language. However, when confined to a small setting, would it be possible to develop an NLI that through repeated iterations can reach perfect understanding? The chosen setting was Risk and was created in Java. The game used Regex to detect certain key elements in the input and interpreted them accordingly. User studies were used to determine the accuracy of the NLI and based on the incorrectly interpreted input the game was improved upon. This was iterated three times. The conclusion was that, while it would be difficult to have the NLI reach a completely perfect understanding, it is possible to achieve precision close to that.

Knowing whether you can implement something complex in a simple way in your application is always of interest. A natural language interface is some- thing that could theoretically be implemented in a lot of applications but the complexity of most natural language processing algorithms is a limiting factor. The problem explored in this paper is whether a simpler algorithm that doesn’t make use of convoluted statistical models and machine learning can be good enough. We implemented two algorithms, one utilizing Spotify’s own search and one with a more accurate, o✏ine search. With the best precision we could muster being 81% at an average of 2,28 seconds per query this is not a viable solution for a complete and satisfactory user experience. Further work could push the performance into an acceptable range.