One of the most challenging aspects of learning a foreign language is learning to converse in that language. While large language models (LLMs) have made it feasible to build generative AI chatbots that simulate conversation, creating an AI-powered language tutor that is both pedagogically effective and engaging for learners requires solving a host of additional problems. At Duolingo, we have developed an AI conversational tutor embedded in our platform, featuring a character named Lily, who helps users practice real-world conversations in their target language. In building this system, we tackled multiple challenges: adapting dialogue to each learner's proficiency level, sustaining personalized and coherent interactions across sessions, maintaining consistent character-driven personality, and designing a structure that supports both guided and learner-initiated topics. Our solution integrates a three-party conversational architecture, a persistent memory mechanism to retain prior interactions, and real-time conversation evaluation to dynamically adjust to the learner's input. This talk will highlight how generative AI, when coupled with rigorous feedback loops and thoughtful design, can significantly enhance the language learning experience.

Opinionated social media such as product reviews are now widely used by individuals and organizations for their decision making. However, due to the reason of profit or fame, people try to game the system by opinion spamming (e.g., writing fake reviews) to promote or demote some target products. For reviews to reflect genuine user experiences and opinions, such spam reviews should be detected. Prior works on opinion spam focused on detecting fake reviews and individual fake reviewers. However, a fake reviewer group (a group of reviewers who work collaboratively to write fake reviews) is even more damaging as they can take total control of the sentiment on the target product due to its size. This paper studies spam detection in the collaborative setting, i.e., to discover fake reviewer groups. The proposed method first uses a frequent itemset mining method to find a set of candidate groups. It then uses several behavioral models derived from the collusion phenomenon among fake reviewers and relation models based on the relationships among groups, individual reviewers, and products they reviewed to detect fake reviewer groups. Additionally, we also built a labeled dataset of fake reviewer groups. Although labeling individual fake reviews and reviewers is very hard, to our surprise labeling fake reviewer groups is much easier. We also note that the proposed technique departs from the traditional supervised learning approach for spam detection because of the inherent nature of our problem which makes the classic supervised learning approach less effective. Experimental results show that the proposed method outperforms multiple strong baselines including the state-of-the-art supervised classification, regression, and learning to rank algorithms.

Given a water distribution network, where should we place sensors toquickly detect contaminants? Or, which blogs should we read to avoid missing important stories?.

How do blogs cite and influence each other? How do such links evolve? Does the popularity of old blog posts drop exponentially with time? These are some of the questions that we address in this work. Blogs (weblogs) have become an important medium of information because of their timely publication, ease of use, and wide availability. In fact, they often make headlines, by discussing and discovering evidence about political events and facts. Often blogs link to one another, creating a publicly available record of how information and influence spreads through an underlying social network. Aggregating links from several blog posts creates a directed graph which we analyze to discover the patterns of information propagation in blogspace, and thereby understand the underlying social network. Here we report some surprising findings of the blog linking and information propagation structure, after we analyzed one of the largest available datasets, with 45,000 blogs and ≈ 2.2 million blog-postings. Our analysis also sheds light on how rumors, viruses, and ideas propagate over social and computer networks.

Weblogs and message boards provide online forums for discussion that record the voice of the public. Woven into this mass of discussion is a wide range of opinion and commentary about consumer products. This presents an opportunity for companies to understand and respond to the consumer by analyzing this unsolicited feedback. Given the volume, format and content of the data, the appropriate approach to understand this data is to use large-scale web and text data mining technologies.This paper argues that applications for mining large volumes of textual data for marketing intelligence should provide two key elements: a suite of powerful mining and visualization technologies and an interactive analysis environment which allows for rapid generation and testing of hypotheses. This paper presents such a system that gathers and annotates online discussion relating to consumer products using a wide variety of state-of-the-art techniques, including crawling, wrapping, search, text classification and computational linguistics. Marketing intelligence is derived through an interactive analysis framework uniquely configured to leverage the connectivity and content of annotated online discussion.

This paper presents a new challenge for Web monitoring tools: to build a system that can monitor entire web sites effectively. Such a system could potentially be used to discover "silent news" hidden within corporate web sites. Examples of silent news include reorganizations in the executive team of a company or in the retirement of a product line. ChangeDetector, an implemented prototype, addresses this challenge by incorporating a number of machine learning techniques. The principal backend components of ChangeDetector all rely on machine learning: intelligent crawling, page classification and entity-based change detection. Intelligent crawling enables ChangeDetector to selectively crawl the most relevant pages of very large sites. Classification allows change detection to be filtered by topic. Entity extraction over changed pages permits change detection to be filtered by semantic concepts, such as person names, dates, addresses, and phone numbers. Finally, the front end presents a flexible way for subscribers to interact with the database of detected changes to pinpoint those changes most likely to be of interest.

In this paper we present a second generation URL monitoring tool which enables the collaborative evaluation of URL content changes. In our implementation, a document monitoring agent works alongside a recommender system. Using information provided by the monitoring agent, the collaborative system alerts users when documents they are monitoring have changed. The monitoring agent provides automatic evaluation of the nature of the change. Users, however, add subjective evaluations; one user's effort informs all others monitoring the same URL. Based on these subjective evaluations, the collaborative system can filter the changed URLs, providing customized notifications for each user based on individual preferences. In this paper, we describe the implemented system and usage results.

This paper describes a new software agent, the community search assistant, which recommends related searches to users of search engines. The community search assistant enables communities of users to search in a collaborative fashion. All queries submitted by the community are stored in the form of a graph. Links are made between queries that are found to be related. Users can peruse the network of related queries in an ordered way: following a path from a first cousin, to a second cousin to a third cousin, etc. to a set of search results. The first key idea behind the use of query graphs is that the determination of relatedness depends on the documents returned by the queries, not on the actual terms in the queries themselves. The second key idea is that the construction of the query graph transforms single user usage of information networks (e.g. search) into collaborative usage: all users can tap into the knowledge base of queries submitted by others.

The promise of workflow solutions for coordinating organizational processes is currently being obscured by strong criticism of the rigidity of their work representations.This rigidity arises in part from viewing work processes as unfolding along a single line of temporally chained activities.In reality, work evolves both horizontally, in the cooperation of causally unrelated, but information-sharing tasks, and vertically, in the coordination of causally-dependent activities.In this paper, we present our process modeling approach which (1) views documents and tasks as duals of each other, capturing horizontal cooperation and (2) exploits constraints to express the SOP dependencies among related activities and documents within the framework of generative rule-based grammars for processes.thus handling vertical coordination.