Dual Dynamic Memory Network for End-to-End Multi-turn Task-oriented Dialog Systems

End-to-end task-oriented dialogue systems rely heavily on an understanding of the dialog history. This often faces the challenge of inferring which dialog history information is critical to generating responses. In this paper, we address this challenge by leveraging a dialog history manager component that dynamically focuses on dialog history memory. It performs multiple add and forget operations by fusing an enhanced entity representation of dialog history and Knowledge Base (KB) information as queries, remembering entities relevant to responses and filtering out unimportant information. Experimental results on an open task-oriented dialogue dataset show that our model outperforms the baseline system in terms of effectiveness and produces contextually consistent responses.

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.

End-to-end Task-oriented spoken dialog systems typically require modeling two types of inputs, namely, the dialog history which is a sequence of utterances and the knowledge base (KB) associated with the dialog history. While modeling these inputs, current state-of-the-art models typically ignore the rich structure in the knowledge graph or its intrinsic association with the dialog history. In this paper, we propose a Flow-to-Graph seq2seq model (FG2Seq) which can effectively encode knowledge by considering inherent structural information of the knowledge graph and latent semantic information from dialog history. Experiments on two publicly available task oriented dialog datasets show that our proposed FG2Seq achieves robust performance on generating appropriate system responses and outperforms the baseline systems.

BackgroundRecent developments in the domain of biomedical knowledge bases (KBs) open up new ways to exploit biomedical knowledge that is available in the form of KBs. Significant work has been done in the direction of biomedical KB creation and KB completion, specifically, those having gene-disease associations and other related entities. However, the use of such biomedical KBs in combination with patients’ temporal clinical data still largely remains unexplored, but has the potential to immensely benefit medical diagnostic decision support systems.ResultsWe propose two new algorithms, LOADDx and SCADDx, to combine a patient’s gene expression data with gene-disease association and other related information available in the form of a KB, to assist personalized disease diagnosis. We have tested both of the algorithms on two KBs and on four real-world gene expression datasets of respiratory viral infection caused by Influenza-like viruses of 19 subtypes. We also compare the performance of proposed algorithms with that of five existing state-of-the-art machine learning algorithms (k-NN, Random Forest, XGBoost, Linear SVM, and SVM with RBF Kernel) using two validation approaches: LOOCV and a single internal validation set. Both SCADDx and LOADDx outperform the existing algorithms when evaluated with both validation approaches. SCADDx is able to detect infections with up to 100% accuracy in the cases of Datasets 2 and 3. Overall, SCADDx and LOADDx are able to detect an infection within 72 h of infection with 91.38% and 92.66% average accuracy respectively considering all four datasets, whereas XGBoost, which performed best among the existing machine learning algorithms, can detect the infection with only 86.43% accuracy on an average.ConclusionsWe demonstrate how our novel idea of using the most and least differentially expressed genes in combination with a KB can enable identification of the diseases that a patient is most likely to have at a particular time, from a KB with thousands of diseases. Moreover, the proposed algorithms can provide a short ranked list of the most likely diseases for each patient along with their most affected genes, and other entities linked with them in the KB, which can support health care professionals in their decision-making.

How to effectively utilize the dialogue history is a crucial problem in multi-turn dialogue generation. Previous works usually employ various neural network architectures (e.g., recurrent neural networks, attention mechanisms, and hierarchical structures) to model the history. However, a recent empirical study by Sankar et al. has shown that these architectures lack the ability of understanding and modeling the dynamics of the dialogue history. For example, the widely used architectures are insensitive to perturbations of the dialogue history, such as words shuffling, utterances missing, and utterances reordering. To tackle this problem, we propose a Ranking Enhanced Dialogue generation framework in this paper. Despite the traditional representation encoder and response generation modules, an additional ranking module is introduced to model the ranking relation between the former utterance and consecutive utterances. Specifically, the former utterance and consecutive utterances are treated as query and corresponding documents, and both local and global ranking losses are designed in the learning process. In this way, the dynamics in the dialogue history can be explicitly captured. To evaluate our proposed models, we conduct extensive experiments on three public datasets, i.e., bAbI, PersonaChat, and JDC. Experimental results show that our models produce better responses in terms of both quantitative measures and human judgments, as compared with the state-of-the-art dialogue generation models. Furthermore, we give some detailed experimental analysis to show where and how the improvements come from.

Whereas research on recombination indicates that the structure of the firm’s knowledge base has an impact on the usefulness of its innovation, it emphasizes that the usefulness of innovation increases when firms’ search spans organizational boundaries. Current literature is limited however in providing insight in how a firm should organize its boundary-spanning search for recombination, given the specific characteristics of its knowledge base. Delineating a continuum of knowledge structures from highly decomposable to non-decomposable, we contribute to extant literature by demonstrating that the knowledge base and in particular its structure also impacts the organizational governance choice for collaborative recombination. In particular we demonstrate on a sample of 108 Biotech firms engaged in 968 contractual agreements and 152 joint ventures with 928 partners that firms with a decomposable knowledge base are more likely to engage in equity joint ventures for collaboration. Furthermore, we demonstrate that when the partner has a decomposable knowledge base, a joint venture is also more likely to be the chosen mode of collaboration. Results imply that the structure of the knowledge base is not only important for the innovation potential of the firm, but also impacts the governance choices firms make for collaborative recombination.

This article examines the importance of the knowledge base in the development of an intelligent algorithmic and software complex of underground water intake facilities. When developing the computerized knowledge base of underground water intake facilities, the management knowledge base, the classification of knowledge bases, the use of conventional knowledge bases for creating expert systems and storing information about the facility will be studied. A knowledge base structure will be developed when creating an intelligent algorithmic and software complex of underground water intake facilities. When organizing the process of forming underground water intake facilities, the development of a knowledge base for the watertight properties of single-layer moisture layers will be studied. During the formation of underground water intake facilities, knowledge is calculated to solve the issues of determining the watertightness of single-layer earthworks.

Large knowledge bases that are intended for applications such as CAD, corporate repositories or process control will have to be shared by multiple users. For these systems to scale up, to give acceptable performance and to exhibit consistent behavior, it is mandatory to synchronize user transactions using a concurrency control algorithm. In this paper, we examine a novel concurrency control policy called Dynamic Directed Graph (or DDG) policy that effectively exploits the rich semantic structure of a knowledge base.

Structured knowledge bases (KBs) are a foundation of many intelligent applications, yet are notoriously incomplete. Language models (LMs) have recently been proposed for unsupervised knowledge base completion (KBC), yet, despite encouraging initial results, questions regarding their suitability remain open. Existing evaluations often fall short because they only evaluate on popular subjects, or sample already existing facts from KBs. In this work, we introduce a novel, more challenging benchmark dataset, and a methodology tailored for a realistic assessment of the KBC potential of LMs. For automated assessment, we curate a dataset called WD-Known, which provides an unbiased random sample of Wikidata, containing over 3.9 million facts. In a second step, we perform a human evaluation on predictions that are not yet in the KB, as only this provides real insights into the added value over existing KBs. Our key finding is that biases in dataset conception of previous benchmarks lead to a systematic overestimate of LM performance for KBC. However, our results also reveal strong areas of LMs. We could, for example, perform a significant completion of Wikidata on the relations nativeLanguage, by a factor of $$\sim $$ 21 (from 260k to 5.8M) at $$82\%$$ precision, and citizenOf by a factor of $$\sim $$ 0.3 (from 4.2M to 5.3M) at 90% precision. Moreover, we find that LMs possess surprisingly strong generalization capabilities: even on relations where most facts were not directly observed in LM training, prediction quality can be high. We open-source the benchmark dataset and code. ( https://github.com/bveseli/LMsForKBC ).

Knowledge-based visual question answering (VQA) involves answering questions that require external knowledge not present in the image. Existing methods first retrieve knowledge from external resources, then reason over the selected knowledge, the input image, and question for answer prediction. However, this two-step approach could lead to mismatches that potentially limit the VQA performance. For example, the retrieved knowledge might be noisy and irrelevant to the question, and the re-embedded knowledge features during reasoning might deviate from their original meanings in the knowledge base (KB). To address this challenge, we propose PICa, a simple yet effective method that Prompts GPT3 via the use of Image Captions, for knowledge-based VQA. Inspired by GPT-3’s power in knowledge retrieval and question answering, instead of using structured KBs as in previous work, we treat GPT-3 as an implicit and unstructured KB that can jointly acquire and process relevant knowledge. Specifically, we first convert the image into captions (or tags) that GPT-3 can understand, then adapt GPT-3 to solve the VQA task in a few-shot manner by just providing a few in-context VQA examples. We further boost performance by carefully investigating: (i) what text formats best describe the image content, and (ii) how in-context examples can be better selected and used. PICa unlocks the first use of GPT-3 for multimodal tasks. By using only 16 examples, PICa surpasses the supervised state of the art by an absolute +8.6 points on the OK-VQA dataset. We also benchmark PICa on VQAv2, where PICa also shows a decent few-shot performance.

In task-oriented dialogue systems, spoken language understanding, or SLU, refers to the task of parsing natural language user utterances into semantic frames. Making use of context from prior dialogue history holds the key to more effective SLU. State of the art approaches to SLU use memory networks to encode context by processing multiple utterances from the dialogue at each turn, resulting in significant trade-offs between accuracy and computational efficiency. On the other hand, downstream components like the dialogue state tracker (DST) already keep track of the dialogue state, which can serve as a summary of the dialogue history. In this work, we propose an efficient approach to encoding context from prior utterances for SLU. More specifically, our architecture includes a separate recurrent neural network (RNN) based encoding module that accumulates dialogue context to guide the frame parsing sub-tasks and can be shared between SLU and DST. In our experiments, we demonstrate the effectiveness of our approach on dialogues from two domains.

Nowadays, a large portion of clinical data only exists in free text. The wide adoption of Electronic Health Records (EHRs) has enabled the increases in accessing to clinical documents, which provide challenges and opportunities for clinical Natural Language Processing (NLP) researchers. Given free-text clinical notes as input, an ideal system for clinical text understanding should have the ability to support clinical decisions. At corpus level, the system should recommend similar notes based on disease or patient types, and provide medication recommendation, or any other type of recommendations, based on patients' symptoms and other similar medical cases. At document level, it should return a list of important clinical concepts. Moreover, the system should be able to make diagnostic inferences over clinical concepts and output diagnosis. Unfortunately, current work has not systematically studied this system. This study focuses on developing and applying methods/techniques in different aspects of the system for clinical text understanding, at both corpus and document level. We deal with two major research questions: First, we explore the question of How to model the underlying relationships from clinical notes at corpus level? Documents clustering methods can group clinical notes into meaningful clusters, which can assist physicians and patients to understand medical conditions and diseases from clinical notes. We use Nonnegative Matrix Factorization (NMF) and Multi-view NMF to cluster clinical notes based on extracted medical concepts. The clustering results display latent patterns existed among clinical notes. Our method provides a feasible way to visualize a corpus of clinical documents. Based on extracted concepts, we further build a symptom-medication (Symp-Med) graph to model the Symp-Med relations in clinical notes corpus. We develop two Symp-Med matching algorithms to predict and recommend medications for patients based on their symptoms. Second, we want to solve the question of How to integrate structured knowledge with unstructured text to improve results for Clinical NLP tasks? On the one hand, the unstructured clinical text contains lots of information about medical conditions. On the other hand, structured Knowledge Bases (KBs) are frequently used for supporting clinical NLP tasks. We propose graph-regularized word embedding models to integrate knowledge from both KBs and free text. We evaluate our models on standard datasets and biomedical NLP tasks, and results showed encouraging improvements on both datasets. We further apply the graph-regularized word embedding models and present a novel approach to automatically infer the most probable diagnosis from a given clinical narrative.

Developing a dialogue system that accomplishes specific tasks through natural conversations with humans is challenging. Currently, although the task-oriented dialogue system constructed by the traditional pipeline model or sequence-to-sequence model has achieved great success, it is still unable to efficiently utilize the external knowledge base to generate high quality dialogue responses. This paper proposes a Memory-to-Sequence framework that uses Memory Neural Network (MemNN) and Long Short Term Memory (LSTM) joint decoding, which can better capture the dependence between system response and knowledge base. Experiments on In-Car Assistant Dataset show that our model significantly outperforms the baseline model and attains the state-of-the-art performance on the two subtasks of the dataset.

Dialogue generation is getting more and more attention in the area of natural language generation, while existing dialogue models focus on the fluency and content-related for reply primarily and ignore its emotional factor. In recent years, Generative Adversarial Networks (GAN) has shown satisfactory results in text generation instead of dialogue generation. In this paper, we propose a dialogue generation framework - EDGAN, which has multiple generators and one multi-class discriminator. Multiple generators are trained to generate response with different emotion labels respectively. The human-generated dialogues and the machine-generated ones are fed to the discriminator. The output about emotion types from the discriminator are used as penalty for each generator to focus on generating its own examples of a specific emotion type accurately. The experimental results on automatic evaluation and manual evaluation demonstrate that our model can generate more high-quality emotional responses than previous baselines.

Predictable and Efficient Memory Management for Composite Events