|
|
Human-computer interaction model for multi-objective recommender systems
Machala, Patrik ; Peška, Ladislav (advisor) ; Lokoč, Jakub (referee)
One of the most developing research fields of information retrieval are recommender systems. They typically try to recommend a few of the most relevant or most suitable items to users from all the candidates when the number of canidates can be in the or- der of thousands or millions. However, it turns out that relevance alone is not enough. Therefore, this work focuses on multi-objective recommender systems using the beyond- relevance objectives. The aim of the thesis is to find out new knowledge about this specific type of recommendation, especially in the connection with the field of HCI, i.e. user and computer interaction that has not been explored much so far. The software output of the work is a web application and a modified recommender system. These two components were used in a user study, where, among other things, we investigated whether users were willing to explicitly set the parameters for a multi- objective recommender system by assigning weights to each of the objectives, compared different variants of metrics for these objectives, mechanisms for assigning weights and different level of detail of texts and visualization of the explanations of the recommen- dations. The results of our experiment show that users perceive the benefit of setting weights for objectives to improve recommendations....
|
|
|
Detection of protein-ligand binding sites using 3D Vision Transformers
Lopuch, Ondrej ; Hoksza, David (advisor) ; Lokoč, Jakub (referee)
Detection of protein-ligand binding sites plays key role not only in understanding of protein function but it also can be used for computer-aided drug design. Improving these detection can lead to faster drug development. In recent years, many machine learning methods were proposed for this task. Nowadays, transformer architecture be- came one of the most prominent one for non-biological data. Its extension for images, vision transformer, became comparable to state-of-the-art algorithms. Moreover, this vision transformer can be expanded into 3D space. The goal of this thesis is to eval- uate possibilities of extending transformers into 3D, for biological data, specifically for protein-ligand binding site detection, exploiting the qualities of attention mechanism. 1
|
|
|
Web plugin for multiple RNA secondary structure visualization
Hercík, Michal ; Hoksza, David (advisor) ; Lokoč, Jakub (referee)
The study of RNA is important to better understand evolution or some diseases. In this thesis, we present a library written in Typescript that offers methods for visual analysis of multiple RNA secondary structures, but preferably work best with two or three structures. For the representation we use radial diagram, which is generated by template-based method. Methods used in the library for analysis use the template-based generation of the radial diagram by mapping diagrams generated from the same template to each other via the template, so that differences and similarities of structures can be observed. 1
|
|
|
Interactive search in image datasets using CLIP neural network
Vopálková, Zuzana ; Lokoč, Jakub (advisor) ; Hoksza, David (referee)
With the growing importance and volume of multimedia data, interactive search sys- tems are essential to help users efficiently search for specific video sequences based on content. One common task is known scene retrieval, where users try to find a particular scene in a large collection of videos. However, the description of a known scene can be subjective, influenced by the perception and experience of individual users, as well as the differences between human and machine perception. In this paper, the effectiveness of an interactive retrieval system combined with image classification generated by a CLIP neural network is investigated to address this problem. V3C datasets and Marine Video Kit are used to verify the effectiveness of the proposed method. Software is also presen- ted that allows data collection for experiments and subsequent evaluation using a web interface. 1
|
|
|
Construction of time-space trajectories from multimodal data
Hrbáček, Matěj ; Skopal, Tomáš (advisor) ; Lokoč, Jakub (referee)
With the growth of public camera recordings and video streams in recent years, there is an increasing need for automatic processing with limited human input. An important part of the process is detecting moving objects in the video and grouping individual detections across video frames into trajectories. This thesis presents a set of algorithms for creating trajectories from object detections while using a configurable analytic model. Presented algorithms are based on the clustering of detections, later even simple trajectories, into complex trajectories by their features, such as a timestamp (frame), bounding rectangle in the video frame and optionally, image crop defined by the bounding rectangle. To present the usage of the generated trajectories, we then introduce methods for further analysis and data extraction. The first method improves the input detections by adding missing detection due to the detector error. The second one is creating a simple semantic description of trajectories to enable further research, such as action analysis or trajectory searching. 1
|
|
|
Fairness in group recommender systems
Maleček, Ladislav ; Peška, Ladislav (advisor) ; Lokoč, Jakub (referee)
The goal of this thesis is to explore the area of group recommender systems with an emphasis on fairness. In the core part of our thesis, we have created a novel aggregation method called Exactly Proportional Fuzzy D'Hondt's Aggrega- tion that works on top of single-user recommender systems. We have evaluated it on five datasets, in three different recommendation scenarios, and with two different types of artificially created groups. The proposed algorithm performed favorably with respect to several fairness metrics while maintaining a reasonable utility of the recommendation. Furthermore, we have created a set of tools to simplify the evaluation pipeline of group recommender systems. The main parts of the pipeline are a dataset downloader, matrix factorizer, and synthetic group creation scripts. We believe these tools may contribute towards more reproducible research in the group recommender systems domain. 1
|
|
|
Extending self-organizing maps with ranking awareness
Park, Kyung Won ; Peška, Ladislav (advisor) ; Lokoč, Jakub (referee)
Title: Extending Self-organizing Maps with Ranking Awareness Author: Kyung Won Park Department: Department of Software Engineering Supervisor: Mgr. Ladislav Peska, Ph.D., Department of Software Engineering Abstract: The self-organizing map (SOM) is a powerful clustering algorithm which takes high- dimensional data as the input and produces a low-dimensional representation of the data. The SOM provides useful insights into the given data by recognizing similar input vectors and clustering them. However, they take into account only the local similarity of the input data, as opposed to relevance (any external ranking). In this paper, we propose two ranking-aware variants of the SOM in an effort to tackle this issue and incorporate evaluation metrics to evaluate our results. Keywords: self-organizing map, relevence feedback, known-item search
|
|
|
Podobnostní vyhledávání obrázků na webu
Grošup, Tomáš ; Lokoč, Jakub (advisor) ; Hoksza, David (referee)
The subject of this bachelor thesis is to design and create a web portal, enabling efficient indexing and content-based searching of images obtained from various free image databases (e.g., results from a keyword-based search engine). The portal provides fast feature extraction technique and for the visual similarity, the signature quadratic form distance is utilized. The search supports various user settings and comparison of their results. Search results can also be presented using a layout based on particle physics, which supports exploration and multi-query.
|
|
|
Presentation Layer For Multimedia Exploration In Multimedia Collections
Macík, Miroslav ; Lokoč, Jakub (advisor) ; Hoksza, David (referee)
Multimedia exploration is addressing the issue of searching within multimedia collections where the data is not annotated, the user is not able to formulate the text query, does not have any example data or wants to get quickly acquainted with the features of the collection. Part of the exploration is also the presentation of the results which is the main contribution of this thesis. The relations between the pictures in the exploration results are transformed into a graph which is visualised through a particle physics model. To visualize a large number of pictures, it is necessary to optimize the calculation. This thesis describes the optimization of both the basic algorithms and their adaptations for the need of multimedia exploration. The calculation layout is available in 2D and 3D space. Powered by TCPDF (www.tcpdf.org)
|
|
|
A fingerpring database
Kubát, Jaroslav ; Babka, Martin (advisor) ; Lokoč, Jakub (referee)
The main goal of this thesis is to provide full description of an implementation of an application that is supposed to be a complex dactyloscopic database tool. The text is divided into parts according to the functional blocks of the application. The parts are namely processing of input fingerprint, the minutiae point analysis, the description of minutiae representation, the database and index structure described with its operations. Usually we describe multiple algorithms for a single task. In this case we provide their comparison. At the end of this thesis, there is a user manual describing the main usage scenarios. Powered by TCPDF (www.tcpdf.org)
|
guest ::
