Few-shot Intent Detection (FSID) is a challenging task in dialogue systems due to the scarcity of available annotated utterances. Although existing few-shot learning approaches have made remarkable progress, they fall short in adapting to the Generalized Few-shot Intent Detection (GFSID) task where both seen and unseen classes are present. A core problem of the simultaneous existence of these two tasks is that limited training samples fail to cover the diversity of user expressions. In this paper, we propose an effective Diversity Features Enhanced Prototypical Network (DFEPN) to enhance diversity features for novel intents by fully exploiting the diversity of known intent samples. Specially, DFEPN generates diversity features of samples in the hidden space via a diversity feature generator module and then fuses these features with original support vectors to get a more suitable prototype vector of each class. To evaluate the effectiveness of our model on both FSID and GFSID tasks, we carry out sufficient experiments on two benchmark intent detection datasets. Results demonstrate that our proposed model outperforms existing state-of-the-art methods and keeps stable performance on both two tasks.

As the smartphone/cellphone market has exploded, the war on which smartphone platform will dominate has become fiercer than ever. In that vain, the goal of this paper is to answer two fundamental questions: What are the adoption trends for smartphones? And how can we estimate the demand for new smartphones? We answer these two questions by collecting a dataset of 3 million subscribers from a nationwide telecom operator. A key aspect of our work is that we have demographic information per user, such as income level, and age, which we correlate with phone usage patterns. Interestingly, we find that in all demographic groups, Android is leading platform top in all age groups and income levels. A key question is whether the “social influence” affects the choice of phone, which we find more pronounced in business plans. Finally, we develop a predictor to infer the phone a user will switch to considering: (a) the type of previous phone, (b) the social influence, and (c) the demographics of the user. Compared with the reference method, our predictor is effective in: (a) reducing the prediction error in number of phones by 1/3, and (b) in the case of minimizing phone costs, the monetary cost by half. Apart from its interest in observations, our work could help telecom operator forecast their inventory more accurately by pointing to the right properties to consider.

How do people use phone calls and text messages for their communication needs? Most studies so far study each mode of communication in isolation. Here, we study the interplay of multi-modal communications. We analyze more than a billion call and text records from a Chinese city and San Francisco Area between 2007 and 2011. First, we provide some definitions towards a framework for analyzing multi-modal communications. Then,we study the relationship of the two communication modes and quantify several aspects of correlation and inference. For a communicating pair, we find that the existence of texting during the weekend is the strongest indicator that the pair will communicate at other times with texts or calls. We compare the behavior between China and the U.S. and we find several similarities and differences. For example, we find evidence of an after-lunch siesta among Chinese users. Finally, we study the evolution of the two modes over time. We find that texting has taken over in sheer number of ”events” by flipping the number of calls over that of texts from 2: 1 in 2007 to 1:2 in 2011.

Homophily refers to the phenomenon where people who are socially-connected share many characteristics including demographic and behavioral properties. The goal of this paper is to see whether homophily exists in call networks and if so, to what degree we can infer a cellphone user's demographic properties by knowing the demographic information of the people that s/he talks to. We focus on three types of demographic information: a) home location, b) age group, and c) income level. The novelty is two-folds. First, we use both communication metrics and structural properties of call graphs to identify those “important” friends for each user with whom (s)he is most likely to be in homophily. Second, we assess the importance of different time slices such as weekdays, or nights and weekends for capturing different user relationships. We conduct our study on a real data trace with 20M subscribers during one month from a nationwide cellular carrier. Our first contribution is that we quantify the extent of homophily on the call graph and identify the correlations between homophily and communication and structural features. As a second contribution, we develop effective methods to infer demographic information for a cellular user using linear regression to select the most homophily-like friend of her/him. We find that we can predict home location within 20km radius with 80% accuracy, and age group and income level with 78% and 72% accuracy, respectively.

Detection of execution anomalies is very important for the maintenance, development, and performance refinement of large scale distributed systems. Execution anomalies include both work flow errors and low performance problems. People often use system logs produced by distributed systems for troubleshooting and problem diagnosis. However, manually inspecting system logs to detect anomalies is unfeasible due to the increasing scale and complexity of distributed systems. Therefore, there is a great demand for automatic anomalies detection techniques based on log analysis. In this paper, we propose an unstructured log analysis technique for anomalies detection. In the technique, we propose a novel algorithm to convert free form text messages in log files to log keys without heavily relying on application specific knowledge. The log keys correspond to the log-print statements in the source code which can provide cues of system execution behavior. After converting log messages to log keys, we learn a Finite State Automaton (FSA) from training log sequences to present the normal work flow for each system component. At the same time, a performance measurement model is learned to characterize the normal execution performance based on the log messages' timing information. With these learned models, we can automatically detect anomalies in newly input log files. Experiments on Hadoop and SILK show that the technique can effectively detect running anomalies.

Researches have discovered that rich interactions among entities in nature and human society bring about complex networks with community structures. In this paper, we propose a novel algorithm BiTector (bi-community detector) to mine the overlapping communities in large-scale sparse bipartite networks. We apply the algorithm to various real-world datasets, showing that BiTector can identify the overlapping community structures in the bipartite networks efficiently and effectively.