A Cooperative Memory Network for Personalized Task-oriented Dialogue Systems with Incomplete User Profiles

User profiling is the process of collecting information about a user in order to construct their profile. The information in a user profile may include various attributes of a user such as geographical location, academic and professional background, membership in groups, interests, preferences, opinions, etc. Big data techniques enable collecting accurate and rich information for user profiles, in particular due to their ability to process unstructured as well as structured information in high volumes from multiple sources. Accurate and rich user profiles are important for applications such as recommender systems, which try to predict elements that a user has not yet considered but may find useful. The information contained in user profiles is personal and thus there are privacy issues related to user profiling. In this position paper, we discuss user profiling with big data techniques and the associated privacy challenges. We also discuss the ongoing EU-funded EEXCESS project as a concrete example of constructing user profiles with big data techniques and the approaches being considered for preserving user privacy.

Large language models (LLMs) like ChatGPT have demonstrated the ability to perform a variety of natural language processing (NLP) tasks. However, it's unclear whether ChatGPT can serve as a task-oriented dialogue system. In this paper, we evaluate the impact of ChatGPT on task-oriented dialogue (TOD) systems and perform a comprehensive analysis to learn its benefits and challenges. We find that ChatGPT performs well on relatively simple dialogue understanding tasks such as intent detection and slot filling, but fails to understand complex multi-turn conversations and interact with Knowledge Base(KB) in dialogue state tracking and response generation. Future LLM-based TOD work should pay more attention to (1) incorporating domain knowledge (2) understanding complex instructions (3) modeling long-term memory (4) interacting with external knowledge bases.

Pre-trained encoder-decoder models are widely applied in Task-Oriented Dialog (TOD) systems on the session level, mainly focusing on modeling the dialog semantic information. Dialogs imply structural information indicating the interaction among user utterances, belief states, database search results, system acts and responses, which is also crucial for TOD systems. In addition, for the system acts, additional pre-training and datasets are considered to improve their accuracies, undoubtedly introducing a burden. Therefore, a novel end-to-end TOD system named Winnie is proposed in this paper to improve the TOD performance. First, to make full use of the intrinsic structural information, supervised contrastive learning is adopted to narrow the gap in the representation space between text representations of the same category and enlarge the overall continuous representation margin between text representations of different categories in dialog context. Then, a system act classification task is introduced for policy optimization during fine-tuning. Empirical results show that Winnie substantially improves the performance of the TOD system. By introducing the supervised contrastive and system act classification losses, Winnie achieves state-of-the-art results on benchmark datasets, including MultiWOZ2.2, In-Car, and Camrest676. Their end-to-end combined scores are improved by 3.2, 1.9, and 1.1 points, respectively.

In recent years the use of personalized service provisioning applications has been very popular. However, effective personalization cannot be achieved without accurate user profiles. In literature a number of classification algorithms have been used to classify user related information to create accurate user profiles. Nevertheless, there is lack of comparison of these algorithms with classification accuracy of the user profile information. In our previous work [1], we compared four different classification algorithms which are; Naive Bayesian (NB), Instance-Based Learner (IB1), Bayesian networks (BN) and Lazy Learning of Bayesian Rules (LBR) classifiers. According to our results NB and IB1 classifiers outperformed the BN and LBR classifiers with respect to classification accuracy. In this study we compare the performance of NB, IB1, Classification and Regression Tree (SimpleCART), Naive Bayesian Tree (NBTree), Iterative Dichotomister Tree (Id3), J48 -a version of C4.5- and Sequential Minimal Optimization (SMO) algorithms with large user profile data. This study is aimed to find the best classification algorithm for user profiling process.Our simulation results show that, in general, the NBTree has the highest classification accuracy performance with the lowest error rate. On the other hand, we also found that the NBTree has one of the highest time requirements to build the classification model. Therefore, NBTree classification algorithm should be favoured over SMO, NB, IB1, J48, SimpleCART and Id3 classifiers in the personalization applications especially when the classification accuracy performance is important.

Task-oriented dialog (TOD) systems typically manage structured knowledge (e.g. ontologies and databases) to guide the goal-oriented conversations. However, they fall short of handling dialog turns grounded on unstructured knowledge (e.g. reviews and documents). In this paper, we formulate a task of modeling TOD grounded on both structured and unstructured knowledge. To address this task, we propose a TOD system with hybrid knowledge management, HyKnow. It extends the belief state to manage both structured and unstructured knowledge, and is the first end-to-end model that jointly optimizes dialog modeling grounded on these two kinds of knowledge. We conduct experiments on the modified version of MultiWOZ 2.1 dataset, where dialogs are grounded on hybrid knowledge. Experimental results show that HyKnow has strong end-to-end performance compared to existing TOD systems. It also outperforms the pipeline knowledge management schemes, with higher unstructured knowledge retrieval accuracy.

Task-Oriented Dialogue (TOD) systems are designed to carry out specific tasks by tracking dialogue states and generating appropriate responses to help users achieve defined goals. Recently, end-to-end dialogue models pre-trained based on large datasets have shown promising performance in the conversational system. However, they share the same parameters to train tasks of the dialogue system (NLU, DST, NLG), so debugging each task is challenging. Also, they require a lot of effort to fine-tune large parameters to create a task-oriented chatbot, making it difficult for non-experts to handle. Therefore, we intend to train relatively lightweight and fast models compared to PLM. In this paper, we propose an End-to-end TOD system with Task-Optimized Adapters which learn independently per task, adding only small number of parameters after fixed layers of pre-trained network. We also enhance the performance of the DST and NLG modules through reinforcement learning, overcoming the learning curve that has lacked at the adapter learning and enabling the natural and consistent response generation that is appropriate for the goal. Our method is a model-agnostic approach and does not require prompt-tuning as only input data without a prompt. As results of the experiment, our method shows competitive performance on the MultiWOZ benchmark compared to the existing end-to-end models. In particular, we attain state-of-the-art performance on the DST task of 2.2 dataset.

Bodybuilding athletes often face challenges in maintaining a diet that aligns with their training and performance goals. This project addresses that issue by developing the Diet Advisor app—a Flutter-based mobile application that offers personalized diet and exercise recommendations based on user profiles and daily activities. The system integrates three core components: user profiling, AI-driven recommendations, and data visualization. Key challenges included accurate data plotting, exercise tracking, and handling incomplete user profiles. These were addressed through adaptive UI design and fallback mechanisms [10]. Two experiments tested the AI's reliability in calculating BMR and generating diet plans. Results demonstrated strong accuracy and high practical value, with no significant discrepancies in BMR values and consistently high ratings for dietary recommendations. Compared to existing AI-based systems, this app provides superior precision and adaptability for bodybuilding needs [1]. Its user-friendly design and scientific basis make it a promising tool for promoting performance-oriented and sustainable nutritional practices.

User profile data (for example, age and sex) is usually self-reported by users, so it is prone to human errors or biases. For example, a user can be reluctant to provide a company with private information such as his/her actual age upon subscription, thus the user either does not fill in the age column or put in some random numbers to avoid unwanted privacy intrusion. However, inaccurate or uncertain user profile data undermines the integrity of a company's marketing or operational intelligence. Targeting customers based on uncertain user profile data will not be as effective as targeting customers based on accurate user profile data. Thus companies perform preprocessing on user profile data as part of effort to maintain the accuracy of their user profile data. This paper presents a study of preprocessing uncertain user profile data based on a proposed simple collaborative learning algorithm. We demonstrate that a user's accurate profile information can be inferred from profile information of the user's social network neighbors. Particularly, we address the issue of how a communication service company can verify whether a user's reported age is true or not. We implement a simple collaborative learning algorithm using mobile network data. The dataset contains anonymized user data from a large Korean mobile company, capturing 174,071 users' demographic profiles and their communication histories. To construct a mobile social network among users, we collect 3G voice call histories including 561,787 unique call receivers who belong to the same service carrier. Results reveal that the prediction accuracy of the proposed method based on voice network data is 97% which is very high compared to 53%, the best accuracy by among competing methods and indicates that our method effectively detects users with great discrepancy between self-reported age and actual age.

Recently, collaborative tagging has become more and more popular in the Web2.0 community, since tags in these Web2.0 systems reflect the specific content features of the resources. This paper presents a recommender for scientific literatures based on semantic concept similarity computed from the collaborative tags. User profiles and item profiles are presented by these semantic concepts, and neighbor users are selected using collaborative filtering. Then, content-based filtering approach is used to generate recommendation list from the papers these neighbor users tagged. The evaluation is carried out on a dataset crawled from CiteULike, with satisfied experiment results.KeywordsRecommenderSemantic conceptWeb2.0Collaborative tag

With the emerging of SOA, Web Services are used frequently to build software applications in a loosely coupled manner. A Web Service is the wide popular implementation of SOA. It is a software component allowing to expose services via Internet. It insures interactions between heterogeneous applications and systems and is platform independent. With the increasing growth in popularity of Web Services, discovering relevant Web Services becomes a significant challenge. Nevertheless, existing Web Services offer the same result regardless the user profile. Therefore, we are interested in defining an adaptable Web Service architecture. In a previous work we have proposed an UML extension named AWS-UML that describes the different allowed users profiles at the metamodel level according to the MDA principles and standards. As we modified the modeling process of the Web Service and so the resulting implementation code, we need also to extend the service description in order to support the additional adaptation informations. Thus, we introduce in the current paper an extension of the standard WSDL used to describe the service offered functionalities. Our proposal is named AWS-WSDL.

Knowledge networks and recommender systems are especially important for expert finding within organizations and scientific communities. Useful recommendation of experts, however, is not an easy task for many reasons: It requires reasoning about multiple complex networks from heterogeneous sources (such as collaboration networks of individuals, article citation networks, and concept networks) and depends significantly on the needs of individuals in seeking recommendations.Although over the past decade much effort has gone into developing techniques to increase and evaluate the quality of recommendations, personalizing recommendations according to individuals' motivations has not received much attention. While previous work in the literature has focused primarily on identifying experts, our focus here is on personalizing the selection of an expert through a principled application of social science theories to model the user's motivation. In this paper, we present an expert recommender system capable of applying multiple theoretical mechanisms to the problem of personalized recommendations through profiling users' motivations and their relations. To this end, we use the Multi-Theoretical Multi-Level (MTML) framework which investigates social drivers for network formation in the communities with diverse goals. This framework serves as the theoretical basis for mapping motivations to the appropriate domain data, heuristic, and objective functions for the personalized expert recommendation.As a proof of concept, we developed a prototype recommender grounded in social science theories, and utilizing computational techniques from social network analysis and representational techniques from the semantic web to facilitate combining and operating on data from heterogeneous sources. We evaluated the prototype's ability to predict collaborations for scientific research teams, using a simple off-line methodology. Preliminary results demonstrate encouraging success while offering significant personalization options and providing flexibility in customizing the recommendation heuristic based on users' motivations. In particular, recommendation heuristics based on different motivation profiles result in different recommendations, and taken as a whole better capture the diversity of observed expert collaboration.

Twitter, Facebook, Pinterest and many other online communities ask their users to populate a location field in their user profiles. The information that is entered into this field has many uses in both industry and academia, with location field data providing valuable geographic context for operators of online communities and playing key roles in numerous research projects. However, despite the importance of location field entries, we know little about how to design location fields effectively. In this paper, we report the results of the first controlled study of the design of location fields in user profiles. After presenting a survey of location field design decisions in use across many online communities, we show that certain design decisions can lead to more granular location information or a higher percentage of users that fill out the field, but that there is a trade-off between granularity and the percent of non-empty fields. We also add context to previous work that found that location fields tend to have a high rate of non-geographic information (e.g. Location: "Justin Bieber's Heart"), showing that this result may be site-specific rather than endemic to all location fields. Finally, we provide evidence that verifying users' location field entries against a database of known-valid locations can eliminate toponym (place name) ambiguity and any non-geographic location field entries while at the same time having little effect on field population rate and granularity.

User modeling for inferring user interests from Online Social Networks (OSNs) such as Twitter has received great attention in the user modeling community with the growing popularity of OSNs. The focus of previous works has been on analyzing user-generated content such as tweets to infer user interests. Therefore, these previous studies were limited to active users who have been actively generating content. On the other hand, with the percentage of passive use of OSNs on the rise, some researchers investigated different types of information about followees (i.e., people that a user is following) such as tweets, usernames, and biographies to infer user interests for passive users who use OSNs for consuming information from followees but who do not produce any content. Although different types of information about followees have been exploited, list memberships (a topical list which other Twitter users can freely add a user into) of followees have not yet been investigated extensively for inferring user interests. In this paper, we investigate list memberships of followees, to infer interest profiles for passive users. To this end, we propose user modeling strategies with two different weighting schemes as well as a refined interest propagation strategy based on previous work. In addition, we investigate whether the information from biographies and list memberships of followees can complement each other, and thus improve the quality of inferred interest profiles for passive users. Results show that leveraging list memberships of followees is useful for inferring user interests when the number of followees is relatively small compared to using biographies of followees. In addition, we found that combining the two different types of information (list memberships and biographies) of followees can improve the quality of user interest profiles significantly compared to a state-of-art method in the context of link recommendations on Twitter.

Task-oriented dialogue systems have become crucial for users to interact with machines and computers using natural language. One of its key components is the dialogue manager, which guides the conversation towards a good goal for the user by providing the best possible response. Previous works have proposed rule-based systems, reinforcement learning, and supervised learning as solutions for correct dialogue management; in other words, select the best response given input by the user. This work explores the impact of dataset quality on the performance of dialogue managers. We delve into potential errors in popular datasets, such as Multiwoz 2.1 and SGD. For our investigation, we developed a synthetic dialogue generator to regulate the type and magnitude of errors introduced. Our findings suggest that dataset inaccuracies, like mislabeling, might play a significant role in the challenges faced in dialogue management.

Building task-oriented dialogue systems has become a topic of interest in the research community and industry. The task-oriented dialogue system is a closed-domain dialogue system that can perform specific tasks for users. The natural language understanding module of a task-oriented dialogue system is crucial because it is related to a task-oriented dialogue system that provides correctional services for users. The natural language understanding module of a task-oriented dialogue system performs two tasks: intent detection and slot filling. The intent detection task can be regarded as a text classification task; a classification model is trained to predict the intention of the user from the user’s input information. The slot filling task can be regarded as a sequence analysis task; a sequence analysis model is trained to predict the details of the user’s intention. In this paper, we proposed a novel model based on a transformer encoder for intent detection and slot filling. It follows the encoder-decoder structure, including a vanilla Transformer encoder, a bidirectional LSTM encoder, a linear classification decoder for intent detection, and a conditional random field decoder for slot filling. The experimental results on two public datasets show that our proposed model outperforms the existing methods based on the Transformer and can be combined with BERT to achieve better intent detection and slot filling results.