Internet Engineering Task Force A. Ovcharenko Internet-Draft 25 July 2023 Intended status: Informational Expires: 26 January 2024 Improving Data Quality through Special Text Tags draft-improving-data-quality-tags-00 Abstract This document proposes the use of special text tags to enhance data quality and improve the understanding of user queries in conversational AI models. By incorporating these tags, models can benefit from additional context and structure during training and inference, leading to more accurate and relevant responses. Status of This Memo This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79. Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet- Drafts is at https://datatracker.ietf.org/drafts/current/. Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress." This Internet-Draft will expire on 26 January 2024. Copyright Notice Copyright (c) 2023 IETF Trust and the persons identified as the document authors. All rights reserved. This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/ license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License. Ovcharenko Expires 26 January 2024 [Page 1] Internet-Draft Improving Data Quality through Special T July 2023 Table of Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 2 2. Motivation . . . . . . . . . . . . . . . . . . . . . . . . . 2 3. Specification . . . . . . . . . . . . . . . . . . . . . . . . 3 3.1. Intent Tagging . . . . . . . . . . . . . . . . . . . . . 3 3.2. Entity Tagging . . . . . . . . . . . . . . . . . . . . . 3 3.3. Contextual Tags . . . . . . . . . . . . . . . . . . . . . 3 3.4. Quality Assessment Tags . . . . . . . . . . . . . . . . . 4 3.5. Emotion or Tone Markers . . . . . . . . . . . . . . . . . 4 4. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 4 5. Security Considerations . . . . . . . . . . . . . . . . . . . 5 6. Interoperability . . . . . . . . . . . . . . . . . . . . . . 5 7. Implementation and Deployment . . . . . . . . . . . . . . . . 5 8. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 5 9. Informative References . . . . . . . . . . . . . . . . . . . 5 Author's Address . . . . . . . . . . . . . . . . . . . . . . . . 6 1. Introduction Conversational AI models often face challenges in data collection and text parsing, impacting their performance and reliability. This proposal aims to address these challenges by introducing special text tags. This approach draws inspiration from related works in natural language processing, information retrieval, and conversational AI. 2. Motivation The motivation behind this proposal is to improve the quality of training data and enhance the understanding of user queries by incorporating special text tags. The idea is influenced by research on intent recognition, entity extraction, and context modeling in natural language understanding. Notable works include: * Previous studies on intent recognition in dialogue systems have explored the use of intent tags to improve the accuracy of responses[intent-recognition]. * Named Entity Recognition (NER) techniques have been widely studied and applied in information extraction tasks. These approaches inspire the entity tagging component proposed in this study[gibbs-sampling]. * Research on dialogue modeling has emphasized the importance of context and sequential information in generating coherent responses. Contextual tags introduced in this proposal draw inspiration from these studies[contextual-understanding]. Ovcharenko Expires 26 January 2024 [Page 2] Internet-Draft Improving Data Quality through Special T July 2023 3. Specification 3.1. Intent Tagging Intent tags are used to label the intent or purpose of user queries, providing guidance to the model in generating more contextually appropriate responses. * [intent-def]: For queries seeking definitions of terms. * [intent-comp]: For queries comparing two or more entities. * [intent-ex]: For queries requesting examples or instances. * [intent-steps]: For queries seeking step-by-step instructions. * [intent-adv-disadv]: For queries exploring the pros and cons of a topic. 3.2. Entity Tagging Entity tags are used to identify and label specific entities within the text, improving the model's understanding of user queries related to those entities. * [entity-person]: For queries related to people or individuals. * [entity-organization]: For queries related to organizations or companies. * [entity-location]: For queries related to specific locations. * [entity-date]: For queries related to dates or time. * [entity-product]: For queries related to products or items. 3.3. Contextual Tags Contextual tags mark contextual information, providing cues for maintaining a coherent and context-aware conversation. * [context-background]: For providing background information or context. * [context-constraints]: For indicating limitations or constraints. * [context-previous-query]: For referring to a previous user query or conversation context. Ovcharenko Expires 26 January 2024 [Page 3] Internet-Draft Improving Data Quality through Special T July 2023 * [context-next-steps]: For suggesting the next steps in a process or task. * [context-clarification]: For seeking clarification or additional details. 3.4. Quality Assessment Tags Quality assessment tags help identify the quality or reliability of information, enabling the model to generate more cautious and reliable responses. * [qa-biased]: Indicating biased information. * [qa-unverified]: Denoting information that is not verified or lacks credibility. * [qa-misleading]: Highlighting information that may be misleading or deceptive. * [qa-outdated]: Identifying information that is outdated or no longer accurate. * [qa-fact-check-needed]: Flagging information that requires fact- checking. 3.5. Emotion or Tone Markers Emotion or tone markers indicate the emotional or tonal aspects of the text, enabling the model to generate more appropriate and empathetic responses. * [tone-positive]: Denoting a positive emotional tone. * [tone-negative]: Indicating a negative emotional tone. * [tone-neutral]: Denoting a neutral or unbiased tone. * [tone-joy]: Indicating a joyful or happy emotion. * [tone-sadness]: Denoting a sad or sorrowful emotion. 4. IANA Considerations This memo includes no request to IANA. Ovcharenko Expires 26 January 2024 [Page 4] Internet-Draft Improving Data Quality through Special T July 2023 5. Security Considerations The security considerations section highlights that implementing special text tags does not introduce inherent security risks. However, it emphasizes the need to ensure secure and privacy- conscious practices during the tagging process and data collection, adhering to existing guidelines[usage-policies]. 6. Interoperability Interoperability is crucial for the widespread adoption of special text tags. This section recognizes the importance of standardization efforts to ensure consistent usage and interpretation of tags across different conversational AI models and platforms. It encourages collaboration with standardization bodies and references existing efforts in the field[caml-dialogue-systems]. 7. Implementation and Deployment The implementation and deployment section discuss the practical aspects of integrating special text tags. It suggests involving human annotators or domain experts to accurately tag training data, modifying training processes to consider the tags, and updating inference systems to interpret and respond to tagged user queries effectively. 8. Conclusion The proposed special text tags offer a structured approach to enrich the training data of conversational AI models. By incorporating these tags, models can improve data quality, enhance understanding of user queries, and generate more accurate and contextually relevant responses. The conclusion section summarizes the potential benefits and encourages further research and experimentation. 9. Informative References [intent-recognition] Chen, M., Xu, Z., Weinberger, K., and O. Chapelle, "Marginalized Denoising Autoencoders for Domain Adaptation", 2012, . Ovcharenko Expires 26 January 2024 [Page 5] Internet-Draft Improving Data Quality through Special T July 2023 [gibbs-sampling] Finkel, J. R., Grenager, T., and C. Manning, "Incorporating Non-local Information into Information Extraction Systems by Gibbs Sampling", 2005, . [contextual-understanding] Ritter, A., Cherry, C., and B. Dolan, "Data-driven Response Generation in Social Media", 2011, . [usage-policies] OpenAI, "Usage policies", 2021, . [caml-dialogue-systems] Kovasznai, G., Kotropoulos, C., and I. Pitas, "CAML - A Universal Configuration Language for Dialogue Systems", . Author's Address Aleksey Ovcharenko Email: aleksey.ovcharenko@gmail.com Ovcharenko Expires 26 January 2024 [Page 6]