Citation Context

... ! "# $ h "# $ h % "$ h !s&s Figure 1: Sequence-Level Uncertain Data Model To our knowledge, =-=[1]-=- is the only work that studies mining sequential patterns from uncertain data. However, this work adopts expected support as the measurement of pattern frequentness, which has some inherent weaknesses...

Citation Context

...nnot detect temporal community trend based outliers because they do not involve any notion of communities. In the first step of discovering normal trends, probabilistic sequential patternminingmethods=-=[9,21]-=-canbeusedtoextracttemporalpatterns,however the patterns detected by these methods are “hard patterns” which are incapable of capturing subtle trends, as discussed in Sec. 2. We propose to tackle this ...

Citation Context

... A recent survey for comparing these two measures and analyzing their relationships is given in [27]. For the problem of uncertain sequential patterns, to the best of our knowledge, the approaches in =-=[28]-=-, [29] are the only studies in the literature. In [28], the authors measure the frequentness of a pattern in uncertain event databases based on its expected support, but this may sometimes lead to the...

Citation Context

... a search term “Tiger” could be disambiguated as 〈(Animal, 0.5), (Sports Personality, 0.3), (Airplane, 0.1), . . .〉. Our Contributions. The problem of SPM in probabilistic databases was introduced in =-=[16]-=-. Our main contributions are as follows. (1) We formalize the notion of event-level uncertainty, provide a possibleworlds semantics, and contrast it with source-level uncertainty. From the possiblewor...

Citation Context

...lity (c.f. Section 2). A recent survey comparing expected support and frequentness probability is given in [22]. For the problem of uncertain sequential patterns, there is less work in the literature =-=[18, 28, 16, 23]-=-. In [18], the authors measure the frequentness of a pattern in uncertain event databases based on the expectation measurement, but this may sometimes lead to the loss of interesting patterns [6, 21]....

Citation Context

...s to determine the frequent ones, most early sequential pattern mining methods are based on A-prior algorithm[2] where the main idea is that the sub-sequences of frequent patterns are also frequent. So we can generate candidate longer frequent patterns from the shorter ones which can substantially reduce the search space to be examined. However, those methods can not deal with the situations where the sequences are uncertain. Consider the amount of noise in our data, it is important for us to use soft-quantization which means that standard mining methods do not apply. There are some work [33] [13] which address the task when the database is uncertain. But the uncertainty in these work [33] [13] occurs at different levels as ours. Muhammad et al. [13] propose a method to mine sequential patterns when there is uncertainty about which sequence an item is associated with (but this is not equivalent to our problem). Zhao et al. [33] solve a similar problem as ours. However, it is impossible to incorporate the gap constraints using their framework. 2640 3. Action Representation We represent an action by a sequence of 3D poses {y(1), · · · , y(m)} where each pose y is a high dimensional vecto...

Citation Context

...ses, measures like expected support [1] and probabilistic frequentness [3] have been used. An expected support based approach for mining sequential patterns in probabilistic databases was proposed in =-=[6]-=-. Here, we focus on probabilistic frequent sequential patterns. Definition 1. Given a probabilistic database D p and its set of possible worlds PW(D p ), the support probability for a sequence s is de...

Citation Context

...t detect temporal community trend based outliers because they do not involve any notion of communities. In the first step of discovering normal trends, probabilistic sequential pattern mining methods =-=[24, 117]-=- can be used to extract temporal patterns, however the patterns detected by these methods are “hard patterns” which are incapable of capturing subtle trends, as discussed in Sec. 4.2. We propose to ta...