1,721,379 research outputs found
Inspecting Deep NLP Models - Innovative Techniques for Explainability, Fairness, and Concept Drift in Natural Language Classification
No full textL'abstract è presente nell'allegato / the abstract is in the attachmen
Multiobjective Combinatorial Optimization with Interactive Evolutionary Algorithms: the case of facility location problems
Full text linkWe consider multiobjective combinatorial optimization problems handled by preference-driven efficient heuristics. They look for the most preferred part of the Pareto front based on some preferences expressed by the user during the process. In general, the Pareto set of efficient solutions is searched for in this case. However, obtaining the Pareto set does not solve the decision problem since one or more solutions, being the most preferred for the user, have to be selected. Therefore, it is necessary to elicit their preferences. What we are proposing can be seen as one of the first structured methodologies in facility location problems to search for optimal solutions taking into account the preferences of the user. To this aim, we use an interactive evolutionary multiobjective optimization procedure called NEMO-II-Ch. It is applied to a real-world multiobjective location problem with many users and many facilities to be located. Several simulations have been performed. The results obtained by NEMO-II-Ch are compared with those obtained by three algorithms knowing the user’s “true” value function that is, instead, unknown to NEMO-II-Ch. They show that in many cases, NEMO-II-Ch finds the best subset of locations more quickly than the methods knowing the whole user’s true preferences
A portfolio approach for the selection and the timing of urban planning projects
No full textThis paper presents a model to support organizations dealing with urban planning decisions. In particular, we deal with the selection and the timing of several projects that can be realised within an urban development project with the aim of optimizing conflicting objectives. The available resources should be allocated in an acceptable way and current and future requirements must be met. In addition, projects can be characterized by qualitative performances and different scenarios can be considered. The model can be handled with different methods. Given its participatory features, it seems that an interactive multiobjective optimization methodology is the most appropriate approach, since the decision maker can express his opinion throughout the resolution process. We present how the model can work through the description of an application based on the execution of the Master Plan of the University of Portsmouth, one of the most fast progressing University in UK. We show the versatility of the model and how it can successfully handle such complex decisions
The Deck-of-cards-based Ordinal Regression method and its application for the development of an ecovillage
Full text linkThis paper presents the deck-of-cards-based Ordinal Regression (DOR), a new multicriteria decision-aiding procedure that conjugates the deck-of-cards method with an ordinal regression approach to define a multicriteria value function representing the preferences of the decision maker (DM). The deck-of-cards method allows the DM to express the ranking order of a set of reference alternatives along with the intensity of preferences between reference alternatives. An ordinal regression procedure is then used to define a multicriteria value function that represents the ranking of the reference alternatives as well as the preference intensity.
This approach can be applied to define value functions with different formulations, such as weighted sum,
additive value, or Choquet integral. The value function thus obtained can be used to comprehensively
evaluate alternatives of a multi-criteria decision problem. The value function provided by DOR can also
be applied to a multi-objective optimisation problem. In this study, we applied DOR to handle urban and
regional planning decisions in which facilities are required to be selected, located, and planned. In particular, we consider the interactions between criteria and synergies between facilities in an enriched version of the so-called space-time model. We applied this methodology to a real-world problem to plan the development of a sustainable ecovillage in the province of Turin (Italy), thus supporting the president of the cooperative owning the ecovillage in his decisions regarding which structures to select, where to locate them, and when to plan their realisation
Drift Lens: Real-time unsupervised Concept Drift detection by evaluating per-label embedding distributions
No full textDespite the significant improvements made by deep learning models, their adoption in real-world dynamic applications is still limited.
Concept drift is among the open issues preventing the widespread exploitation of deep learning models in real-life settings. The dynamic world changes very quickly, and the collected data drifts accordingly.
Prediction models, usually trained on static historical data, should be promptly re-trained in case of new real-time drifted data distributions.
Although some drift detection methodologies have been proposed over the years, different issues are still open since state-of-the-art solutions show limited effectiveness and efficiency.
This paper proposes DRIFT LENS , a novel real-time unsupervised per-label drift detection methodology based on embedding distribution distances in deep learning models.
The preliminary experiments performed on a transformer-based model fine-tuned for topic text classification show promising results in drift detection accuracy, drift characterization, and efficient execution time to support real-time concept drift detection
A general space-time model for combinatorial optimization problems (and not only)
Full text linkWe consider the problem of defining a strategy consisting of a set of facilities taking into account also the location where they have to be assigned and the time in which they have to be activated. The facilities are evaluated with respect to a set of criteria. The plan has to be devised respecting some constraints related to different aspects of the problem such as precedence restrictions due to the nature of the facilities. Among the constraints, there are some related to the available budget. We consider also the uncertainty related to the performances of the facilities with respect to considered criteria and plurality of stakeholders participating to the decision. The considered problem can be seen as the combination of some prototypical operations research problems: knapsack problem, location problem and project scheduling. Indeed, the basic brick of our model is a variable xilt which takes value 1 if facility i is activated in location l at time t, and 0 otherwise. Due to the conjoint consideration of a location and a time in the decision variables, what we propose can be seen as a general space-time model for operations research problems. We discuss how such a model permits to handle complex problems using several methodologies including multiple attribute value theory and multiobjective optimization. With respect to the latter point, without any loss of the generality, we consider the compromise programming and an interactive methodology based on the Dominance-based Rough Set Approach. We illustrate the application of our model with a simple didactic example
Optimization of multiple satisfaction levels in portfolio decision analysis
No full textWe consider a portfolio decision problem in which a set of projects forming a portfolio has to be selected taking into account multiple evaluation criteria and some constraints related to the limited resources (e.g., available budget). Traditionally, such a problem has been approached by Multiple Attribute Value Theory (MAVT) with the aim of maximizing the sum of values associated with the projects included in the selected portfolio. Using MAVT, one represents preferences on the individual projects, and a value of a portfolio is just an aggregate of values of the component projects. This linear value approach does not explicitly account for portfolio balance requirements, raising the risk of selecting a portfolio which is, e.g., composed of projects with good evaluations on the same criterion or on the same small subset of criteria. Thus, we propose a different approach that enables the Decision Maker (DM) to control the distribution of good evaluations on different criteria over the projects composing a portfolio. With this aim, for each criterion we fix a certain number of reference levels corresponding to the qualitative satisfaction degrees. The number of projects entering a portfolio and attaining each of these levels becomes an objective to be maximized. To solve thus formulated multi-objective optimization problem, we use Dominance-based Rough Set Approach (DRSA). The DM is expected to point out some prospective portfolios in a current sample of non-dominated portfolios. DRSA represents the DM's preferences with a set of decision rules induced from such indirect preference information. Their use permits to progressively focus the search on the part of the non-dominated portfolios that satisfy the DM's preferences in the best way
Decoding Narratives: Towards a Classification Analysis for Stereotypical Patterns in Italian News Headlines
Full text linkMedia headlines shape our initial interpretation of news, framing narratives that influence societal engagement with political and social issues. Yet, they often rely on sensationalism and bias to capture readers’ attention. In this paper, we aim to uncover distinct patterns in Italian headline composition, examining how language and framing vary across political leanings. We analyze a dataset of daily Italian newspaper articles from two outlets with opposing political perspectives, anonymized as Newspaper A and Newspaper B. Our study encompasses the entire set of news and a subset of topics (n = 8) likely to contain stereotypes or clickbait headlines identified using a Large Language Model. Our methodology combines (1) a lexicometric analysis to identify characteristic words of each newspaper, and (2) the training of an accurate deep learning classifier (F1 = 0.84) to learn specific patterns for categorizing headlines into these two perspectives and leveraging explainability techniques to extract and interpret these patterns. Our analysis reveals distinct tonal differences between the two newspapers: Newspaper A generally adopts a more balanced and nuanced approach, while Newspaper B often favors a more direct and sometimes provocative style, especially regarding topics like immigration and social justice. Additionally, Newspaper B’s headlines tend to be brief and punchy, in contrast to the longer, more detailed ones from Newspaper A. Despite these tonal differences, both outlets exhibit similar stereotypical patterns in their coverage, such as consistently emphasizing nationality and group distinctions in ways that can reinforce social stereotypes. This shared tendency suggests that, although their narrative strategies differ, both outlets could contribute to a broader pattern of stereotype reinforcement
- …
