Amazon Technologies, Inc.

Jason Roselander

Brandon William Porter

Logan Luyet Dillard

Terrence R Nightingale

Olga I Kuznetsova

Disclosed are various embodiments for extracting an excerpt from a representative review of an item, such as an item available for purchase in an electronic commerce system. Attributes or categories used in reviews of an item may be identified and ranked according to consumer preference. Upon ranking the categories, an excerpt may be extracted from a review corresponding to a ranked one of the attributes or categories. The excerpt may be identified and extracted if a number of reviews for an item exceeds a threshold quantity as it may be impractical for a user to read every review written about the item.

Automatic review excerpt extraction

Disclosed are various embodiments for extracting a representative review excerpt from a plurality of reviews associated with an item available for purchase in an electronic commerce system. A review extractor identifies a lexicon appearing in the reviews and associates each of the reviews with at least one category where the categories are based on the identified common lexicon. The review extractor then ranks the categories according to a ranking of attributes specified by consumer preferences. Upon ranking the categories, the review extractor then extracts a representative review excerpt based on a correlation between the number of reviews associated with each category and the ranking of that category as indicated by the consumer preferences.

Techniques are described for automatically determining the structure of multiple related Web pages based at least in part on human-supplied information, such as in at least some situations by automated operations of a website structure determination system to determine a common structure of at least some Web pages of a target Website of interest. In at least some such situations, the techniques include identifying structural elements or other structural locations within the common structure of multiple related Web pages that correspond to one or more types of information of interest. Once such structural locations are identified for one or more types of information of interest, information of those types may then be automatically retrieved from various Web pages of the target Website using those identified structural locations, even if those Web pages have not been previously manually reviewed.

Growing community of inventors

Luxembourg, Luxembourg

Jason Roselander