Investigating the role of in-situ user expectations in Web search

Pre-adoption expectations often serve as an implicit reference point in users’ evaluation of information systems and are closely associated with their goals of interactions, behaviors, and overall satisfaction. Despite the empirically confirmed impacts, users’ search expectations and their connections to tasks, users, search experiences, and behaviors have been scarcely studied in the context of online information search. To address the gap, we collected 116 sessions from 60 participants in a controlled-lab Web search study and gathered direct feedback on their in-situ expected information gains (e.g., number of useful pages) and expected search efforts (e.g., clicks and dwell time) under each query during search sessions. Our study aims to examine (1) how users’ pre-search experience, task characteristics, and in-session experience affect their current expectations and (2) how user expectations are correlated with search behaviors and satisfaction. Our results with both quantitative and qualitative evidence demonstrate that: (1) user expectation is significantly affected by task characteristics, previous and in-situ search experience; (2) user expectation is closely associated with users’ browsing behaviors and search satisfaction. The knowledge learned about user expectation advances our understanding of users’ search behavioral patterns and their evaluations of interaction experience and will also facilitate the design, implementation, and evaluation of expectation-aware user models, metrics, and information retrieval (IR) systems.     

Personalized fashion outfit generation with user coordination preference learning

This paper focuses on personalized outfit generation, aiming to generate compatible fashion outfits catering to given users. Personalized recommendation by generating outfits of compatible items is an emerging task in the recommendation community with great commercial value but less explored. The task requires to explore both user-outfit personalization and outfit compatibility, any of which is challenging due to the huge learning space resulted from large number of items, users, and possible outfit options. To specify the user preference on outfits and regulate the outfit compatibility modeling, we propose to incorporate coordination knowledge in fashion. Inspired by the fact that users might have coordination preference in terms of category combination, we first define category combinations as templates and propose to model user-template relationship to capture users’ coordination preferences. Moreover, since a small number of templates can cover the majority of fashion outfits, leveraging templates is also promising to guide the outfit generation process. In this paper, we propose Template-guided Outfit Generation (TOG) framework, which unifies the learning of user-template interaction, user–item interaction and outfit compatibility modeling. The personal preference modeling and outfit generation are organically blended together in our problem formulation, and therefore can be achieved simultaneously. Furthermore, we propose new evaluation protocols to evaluate different models from both the personalization and compatibility perspectives. Extensive experiments on two public datasets have demonstrated that the proposed TOG can achieve preferable performance in both evaluation perspectives, namely outperforming the most competitive baseline BGN by 7.8% and 10.3% in terms of personalization precision on iFashion and Polyvore datasets, respectively, and improving the compatibility of the generated outfits by over 2%.     

Search task success evaluation by exploiting multi-view active semi-supervised learning

Search task success rate is an important indicator to measure the performance of search engines. In contrast to most of the previous approaches that rely on labeled search tasks provided by users or third-party editors, this paper attempts to improve the performance of search task success evaluation by exploiting unlabeled search tasks that are existing in search logs as well as a small amount of labeled ones. Concretely, the Multi-view Active Semi-Supervised Search task Success Evaluation (MA4SE) approach is proposed, which exploits labeled data and unlabeled data by integrating the advantages of both semi-supervised learning and active learning with the multi-view mechanism. In the semi-supervised learning part of MA4SE, we employ a multi-view semi-supervised learning approach that utilizes different parameter configurations to achieve the disagreement between base classifiers. The base classifiers are trained separately from the pre-defined action and time views. In the active learning part of MA4SE, each classifier received from semi-supervised learning is applied to unlabeled search tasks, and the search tasks that need to be manually annotated are selected based on both the degree of disagreement between base classifiers and a regional density measurement. We evaluate the proposed approach on open datasets with two different definitions of search tasks success. The experimental results show that MA4SE outperforms the state-of-the-art semi-supervised search task success evaluation approach.     

Delphi study of digital libraries

A Delphi study of digital libraries was undertaken in January 1998 to gain a broader understanding of issues related to DLs. Selected experts in the field were identified and presented with a series of 118 opinion statements, organized over a set of 13 questions. The Delphi Planning System, a Web-based software application, was utilized initially. Originally, 218 experts were considered and 33 were invited to participate in the Delphi study. 21 of those 33 initially agreed to participate. 13 general question areas formulated by the project group were forwarded to the invited experts, some of whom withdrew from participation. A review of the 9 expert contributions generated 118 opinion statements across the set of 13 questions. Round 1 of the Delphi resulted in 6 of 118 opinion statements reaching consensus. Round 2 resulted in 83 of the 118 opinion statements reaching some consensus of opinion. Round 3 focused on the 35 opinion statements that had not reached consensus. Analysis of round 3 data revealed that consensus of opinion or stability had been reached on all statements. Results from expert consensus included that: (1) efforts associated with the development of digital libraries are primarily collaborative, (2) an array of expertise is involved in the research and development of DL, (3) a DL has the potential to transform access to digital knowledge records and (4) the primary role of librarians in DL development is an extension of current practice. The names of 40 contributors to digital library research and development were posited by the study participants, with 10 identified as being `experts in DL'.     

Evaluation of a personalized digital library based on cognitive styles: Adaptivity vs. adaptability

Personalization can be addressed by adaptability and adaptivity, which have different advantages and disadvantages. This study investigates how digital library (DL) users react to these two techniques. More specifically, we develop a personalized DL to suit the needs of different cognitive styles based on the findings of our previous work [Frias-Martinez, E., Chen, S. Y., & Liu, X. (2008) Investigation of behavior and perception of digital library users: A cognitive style perspective. International Journal of Information Management]. The personalized DL includes two versions: adaptive version and adaptable version. The results showed that users not only performed better in the adaptive version, but also they perceived more positively to the adaptive version. In addition, cognitive styles have great effects on users’ responses to adaptability and adaptivity. These results provide guidance for designers to select suitable techniques to develop personalized DLs.     

Dual disentanglement of user–item interaction for recommendation with causal embedding

To achieve personalized recommendations, the recommender system selects the items that users may like by learning the collected user–item interaction data. However, the acquisition and use of data usually form a feedback loop, which leads to recommender systems suffering from popularity bias. To solve this problem, we propose a novel dual disentanglement of user–item interaction for recommendation with causal embedding (DDCE). Different from the existing work, our innovation is we take into account double-end popularity bias from the user-side and the item-side. Firstly, we perform a causal analysis of the reasons for user–item interaction and obtain the causal embedding representation of each part according to the analysis results. Secondly, on the item-side, we consider the influence of item attributes on popularity to improve the reliability of the item popularity. Then, on the user-side, we consider the effect of the time series when obtaining users’ interest. We model the contrastive learning task to disentangle users’ long–short-term interests, which avoids the bias of long–short-term interests overlapping, and use the attention mechanism to realize the dynamic integration of users’ long–short-term interests. Finally, we realize the disentanglement of user–item interaction reasons by decoupling user interest and item popularity. We experiment on two real-world datasets (Douban Movie and KuaiRec) to verify the significance of DDCE, the average improvement of DDCE in three evaluation metrics (NDCG, HR, and Recall) compared to the state-of-the-art model are 5.1106% and 4.1277% (MF as the backbone), 3.8256% and 3.2790% (LightGCN as the backbone), respectively.     

Implicit dimension identification in user-generated text with LSTM networks

In the process of online storytelling, individual users create and consume highly diverse content that contains a great deal of implicit beliefs and not plainly expressed narrative. It is hard to manually detect these implicit beliefs, intentions and moral foundations of the writers.We study and investigate two different tasks, each of which reflect the difficulty of detecting an implicit user’s knowledge, intent or belief that may be based on writer’s moral foundation: (1) political perspective detection in news articles (2) identification of informational vs. conversational questions in community question answering (CQA) archives. In both tasks we first describe new interesting annotated datasets and make the datasets publicly available. Second, we compare various classification algorithms, and show the differences in their performance on both tasks. Third, in political perspective detection task we utilize a narrative representation language of local press to identify perspective differences between presumably neutral American and British press.     

How to approximate users’ values while preserving privacy: experiences with using attitudes towards work tasks as proxies for personal value elicitation

Software users have different sets of personal values, such as benevolence, self-direction, and tradition. Among other factors, these personal values influence users’ emotions, preferences, motivations, and ways of performing tasks—and hence, information needs. Studies of user acceptance indicate that personal traits like values and related soft issues are important for the user’s approval of software. If a user’s dominant personal value were known, software could automatically show an interface variant which offers information and functionality that best matches his or her dominant value. A user’s dominant personal value is the one that most strongly influences his or her attitudes and behaviors. However, existing methods for measuring a user’s values are work intensive and/or interfere with the user’s privacy needs. If interface tailoring for very large groups of users is planned, value approximation has to be achieved on a large scale to assign individualized software to all users of the software. Our work focuses on approximating the dominant values of a user with less effort and less impact on privacy. Instead of probing for a user’s values directly, we explore the potential of approximating these values based on the user’s preferences for key tasks. Producing tailored versions of software is a separate topic not in the focus here. In this paper we rather describe a method to identify user values from task preferences and an empirical study of applying parts of this method. We are proposing the method in this paper for the first time except for a preliminary version orally presented at a workshop. The method consists of a research process and an application process. In the research process a researcher has to identify key tasks occurring in a context under investigation which have a relationship to personal values. These key tasks can be used in the application process to approximate the dominant values of new users in a similar context. In this empirical study we show that the research process of our method allows us to determine key tasks which approximate values in the shared context of nursing. The majority of the nurses were found to have one of the three following dominant values: benevolence, self-direction, or hedonism. Data confirmed common expectations: that nurses with the value of benevolence, when compared to all other nurses, had a higher preference for tasks which helped people immediately or improved their circumstances of the treatment. In relation to all other nurses, participants with self-direction disliked tasks which affected their personal freedom, and users with hedonism had a lower preference for tasks which involved physical work and preferred tasks which promised gratification. Our findings advance measurement of personal values in large user groups by asking questions with less privacy concern. However, the method requires substantial efforts during the initial research process to prepare such measurements. Future work includes replicating our method in other contexts and identifying value-dependent tasks for users with other values than the three values our empirical study mainly focused on.     

数字图书馆的生命周期成本模型

从数字图书馆的成本入手,提炼出"数字图书馆的五类成本"并结合国内数字图书馆实际建设案例,构建了数字资源8阶段-6代理生命周期模型,同时,再以前两者为基础构建了"数字图书馆生命周期三维成本模型"。     

Asking Clarifying Questions: To benefit or to disturb users in Web search?

Modern information-seeking systems are becoming more interactive, mainly through asking Clarifying Questions (CQs) to refine users’ information needs. System-generated CQs may be of different qualities. However, the impact of asking multiple CQs of different qualities in a search session remains underexplored. Given the multi-turn nature of conversational information-seeking sessions, it is critical to understand and measure the impact of CQs of different qualities, when they are posed in various orders. In this paper, we conduct a user study on CQ quality trajectories, i.e., asking CQs of different qualities in chronological order. We aim to investigate to what extent the trajectory of CQs of different qualities affects user search behavior and satisfaction, on both query-level and session-level. Our user study is conducted with 89 participants as search engine users. Participants are asked to complete a set of Web search tasks. We find that the trajectory of CQs does affect the way users interact with Search Engine Result Pages (SERPs), e.g., a preceding high-quality CQ prompts the depth users to interact with SERPs, while a preceding low-quality CQ prevents such interaction. Our study also demonstrates that asking follow-up high-quality CQs improves the low search performance and user satisfaction caused by earlier low-quality CQs. In addition, only showing high-quality CQs while hiding other CQs receives better gains with less effort. That is, always showing all CQs may be risky and low-quality CQs do disturb users. Based on observations from our user study, we further propose a transformer-based model to predict which CQs to ask, to avoid disturbing users. In short, our study provides insights into the effects of trajectory of asking CQs, and our results will be helpful in designing more effective and enjoyable search clarification systems.     

Understanding in-context interaction: An investigation into on-the-go mobile search

Recent years have seen a profound change in how most users interact with search engines: the majority of search requests now come from mobile devices, which are used in a number of distracting contexts. This use of mobile devices in various situational contexts away from a desk presents a range of novel challenges for users and, consequently, possibilities for interface improvements. However, there is at present a lack of work that evaluates interaction in such contexts to understand what effects context and mobility have on behaviour and errors and, ultimately, users’ search performance.Through a controlled study, in which we simulate walking conditions on a treadmill and obstacle course, we use a combination of interaction logs and multiple video streams to capture interaction behaviour as participants (n = 24) complete simple search tasks. Using a bespoke tagging tool to analyse these recordings, we investigate how situational context and distractions impact user behaviour and performance, contrasting this with users in a baseline, seated condition. Our findings provide insights into the issues these common contexts cause, how users adapt and how such interfaces could be improved.     

互联网开放式创新社区用户画像构建及用户创新需求识别

本文通过分析互联网开放式创新社区用户参与创新行为，采集用户知识共享的不同特征数据，构建用户画像，从而识别不同类别用户参与平台知识创新的功能和角色。此外，基于用户的异质性，分析不同用户群体的创新需求差异和需求点的分布。通过对小米社区的实证研究识别出了核心用户、积极创新用户、积极社交用户、潜在创意用户、边缘用户这5个类型用户并进行画像，同时从必备型需求、魅力型需求、期望型需求、沉默型需求这4个角度对不同类型用户的创新需求点进行分析，体现了不同类型用户的不同创新需求差异。通过用户画像构建以及用户创新需求识别，达到社区管理与用户之间的高匹配度及用户的创新需求与企业产品创新方向之间的一致性。     

A study of awareness in multimedia search

Awareness of another’s activity is an important aspect of facilitating collaboration between users, enabling an “understanding of the activities of others” (Dourish & Bellotti, 1992). In this paper we investigate the role of awareness and its effect on search performance and behaviour in collaborative multimedia retrieval. We focus on the scenario where two users are searching at the same time on the same task, and via an interface, can see the activity of the other user. The main research question asks: does awareness of another searcher aid a user when carrying out a multimedia search session?To encourage awareness, an experimental study was designed where two users were asked to compete to find as many relevant video shots as possible under different awareness conditions. These were individual search (no awareness), Mutual awareness (where both users could see the other’s search screen), and unbalanced awareness (where one user is able to see the other’s screen, but not vice-versa). Twelve pairs of users were recruited, and the four worst performing TRECVID 2006 search topics were used as search tasks, under four different awareness conditions. We present the results of this study, followed by a discussion of the implications for multimedia information retrieval systems.     

Deciphering cluster representations

There are several recent studies that propose search output clustering as an alternative representation method to ranked output. Users are provided with cluster representations instead of lists of titles and invited to make decisions on groups of documents. This paper discusses the difficulties involved in representing clusters for users’ evaluation in a concise but easily interpretable form. The discussion is based on findings and user feedback from a user study investigating the effectiveness of search output clustering. The overall impression created by the experiment results and users’ feedback is that clusters cannot be relied on to consistently produce meaningful document groups that can easily be recognised by the users. They also seem to lead to unrealistic user expectations.     

同侪影响对用户贡献行为的作用研究———基于网络客观大数据的分析

本文将同侪影响引入在线创新社区的用户行为研究中,从广度和深度两方面考察同侪影响对用户贡献行为的影响,并分析感知收益的中介作用。研究以小米社区MIUI功能与讨论区的创意集市板块为对象构建S-O-R模型,采用6567名用户发布的8830条创意、5.26万条评论和收到的103.36万条评论数据,利用Mplus8.1分析检验,结果发现：同侪影响广度与深度均有利于促进用户贡献行为,综合收益在同侪影响广度、深度与用户贡献行为间起正向中介效应,情感收益仅在同侪影响广度、深度与主动贡献行为间起正向中介效应,而认知收益则在同侪影响深度与反应贡献行为间起负向中介效应。研究拓展了在线网络情境下知识管理与社会学领域的交叉研究,并为在线创新社区社交网络和知识管理提供重要启示。     

基于用户感知的数字信息资源有用性的评价模型构建

对数字信息资源的有用性进行界定,根据用户特定需求建立一套科学、合理的评价指标体系。从经济效益的角度出发,借助层次分析法探讨用户对数字信息资源的有效利用,考察后期服务绩效的目的是在数字信息资源与用户之间构建一个理想的评价体系及和谐的分享环境,进一步提高用户对数字信息资源的利用率。     

A Convolutional Attention Network for Unifying General and Sequential Recommenders

General recommenders and sequential recommenders are two modeling paradigms of recommender. The main focus of a general recommender is to identify long-term user preferences, while the user’s sequential behaviors are ignored and sequential recommenders try to capture short-term user preferences by exploring item-to-item relations, failing to consider general user preferences. Recently, better performance improvement is reported by combining these two types of recommenders. However, most of the previous works typically treat each item separately and assume that each user–item interaction in a sequence is independent. This may be a too simplistic assumption, since there may be a particular purpose behind buying the successive item in a sequence. In fact, a user makes a decision through two sequential processes, i.e., start shopping with a particular intention and then select a specific item which satisfies her/his preferences under this intention. Moreover, different users usually have different purposes and preferences, and the same user may have various intentions. Thus, different users may click on the same items with an attention on a different purpose. Therefore, a user’s behavior pattern is not completely exploited in most of the current methods and they neglect the distinction between users’ purposes and their preferences. To alleviate those problems, we propose a novel method named, CAN, which takes both users’ purposes and preferences into account for the next-item recommendation. We propose to use Purpose-Specific Attention Unit (PSAU) in order to discriminately learn the representations of user purpose and preference. The experimental results on real-world datasets demonstrate the advantages of our approach over the state-of-the-art methods.