Abstract: Disclosed are systems and methods that determine a likelihood that an item page represents a risky item. For example, a machine learning model may be trained to determine a probability score that an item page corresponds to a topic of a plurality of topics based on item textual descriptions determined from the item page and search queries corresponding to the item page. Still further, one or more of a topic risk score, item risk score, cumulative item risk score, and/or a total item risk score may be determined. Each of the risk scores may be indicative of a likelihood that the item page represents a risky item.

Abstract: Systems and techniques are generally described for detecting evasive terms in item listings. In some examples, a list of keywords associated with a first category of specified items may be received. In some examples, text data associated with a first item may be received. A first plurality of tokens representing a corrected version of the text data may be generated using spell correction techniques. In some cases, first flag data indicating that the text data constitutes evasive text may be generated based at least in part on the generation of the first plurality of tokens. The first flag data and a numerical representation of the first plurality of tokens may be input into a binary classifier trained to generate label data indicating whether text data associated with a given item is evasive or non-evasive. The binary classifier may generate label data indicating that the text data associated with the first item is evasive.

Abstract: Systems and techniques are generally described for multi-stage text-based clustering. In various examples, a first data set comprising a first set of elements may be received. Each element of the first set of elements may include text data. A clustering algorithm may be used to determine a first set of clusters of the first set of elements based at least in part on similarities in the text data. A second set of clusters, for each cluster of the first set of clusters, may be determined using the clustering algorithm. A number of resources for labeling the first set of elements may be determined based at least in part on the first set of clusters. The number of resources may be assigned to label clusters of the second set of clusters based at least in part on a number of elements in each cluster of the second set of clusters.