Paypal, Inc.

Avishay Meron

Adam Cohen

Chunmao Ran

Xing Wang

David Stein

Yarden Raiskin

Itzik Levi

Doron Hai-Reuven

Chuanyun Fang

Avihay Nathan

Kun Fu

Raz Alon

Tom Bar Yacov

Amnon Jislin

Anna Livshits-Rozenman

Oded Ben Haim

Tomer Handelman

Shay Elbaz

Dudu Markovitz

Venkatesh Ramanathan

Roy Valler

Shuly Lev-Yehudi

Methods and systems are presented for classifying a particular user account as a fraudulent user account by analyzing links between the user account and two or more known fraudulent user accounts collectively. Attributes of the particular user account are compared against attributes of a plurality of known fraudulent accounts to determine that the particular user account has shared attributes with a first known fraudulent account and a second known fraudulent account. The shared attributes with the first known fraudulent account and the second known fraudulent account are analyzed collectively to determine a risk level for the particular user account. The risk level may indicate a likelihood that the particular user account corresponds to a fraudulent account.

Systems and methods for classifying accounts based on shared attributes with known fraudulent accounts

Placing an event into a particular cluster can allow various inferences about the event. A new payment transaction that looks similar to a previously identified cluster of mostly fraudulent payment transactions, for example, may be higher risk. The present disclosure includes structural data improvements to the way that online clustering of events (which may include web events and not just payment transactions) occurs. A new event can be classified into a particular segment very quickly using feature table searching, which can allow for better decision making when a short timeframe is required (e.g. transaction processing, online advertising, etc.).

Rapid online clustering

Character string analysis and classification can be useful in a variety of contexts, including examining web URLs to determine whether a URL indicates that a user is attempting to take a particular action on an electronic service platform. In some cases, however, URLs or other string data may have 'noise' in them, such as random sub-strings, that prevents a string from being properly classified. Sometimes it may be useful to classify a string into a category, however, and it may be important to do this quickly (e.g. during an active user interaction with a website). Learning tables allowing for O(1) lookup can be established by tokenizing strings and then using probability analysis to eliminate tokens that appear an insufficient number of times. This allows for quick and accurate string classification, which may be useful in numerous circumstances.

Efficient random string processing

A method is provided that includes accessing, by a server provider server of a service provider, a database storing associations between network addresses and locations. Additionally, the method includes determining a subset of the database corresponding to a first network address, each association included in the subset corresponding to an association between the first network address and a respective location. The method also includes in response to determining that the subset of the database satisfies one or more clustering criteria, calculating a representative location corresponding to the first network address, and storing an association between the first network address and the representative location in a second database.

Calculating representative location information for network addresses

Aspects of the present disclosure involve systems, methods, devices, and the like for classifying or indexing websites using an efficient multistage statistical indexing scheme. In one embodiment, a system is introduced that includes a plurality of models for classification. The classification and indexing may be obtained using at least two models designed to retrieve information from a website visited for marketing and assessing a user. In one embodiment, a first model, content-based features are used for the classification and in making predictions of the unlabeled observations. In another embodiment, a second model includes embedded technology-based features for further classification.

System and method for efficient multi stage statistical website indexing

A system and method for generating automatic false positive estimations for website matching is described. Several sets of assets and Uniform Resource Locators (URLs) are aggregated. Each of the several sets of assets is searched across webpage content corresponding to the several URLs to determine matches between the sets of assets and webpage content. One or more false positive estimations is determined, where each of the one or more false positive estimations corresponds to the one or more matches. A combined score is generated based on the one or more false positive estimations.

Automatic false positive estimation for website matching

Systems and methods for machine learning feature processing are disclosed. An example method includes: selecting a predefined number of machine-based transactions; identifying a first set of features based on the predefined number of machine-based transactions; and applying a machine learning feature selection algorithm to identify a second set of features. The second set of features is smaller than the first set of features, and the second set of features includes no more than a predefined number of features. The method also includes constructing a decision tree based on the second set of features, wherein the decision tree has a depth that is less than a predefined depth. The method optionally includes limiting the total number of features in the second set to no greater than the predefined number of features.

Automated machine learning feature processing

Various systems, mediums, and methods herein describe aspects of a proximity and validation based data merging system. The system may access a first data set and a plurality of second data sets that correspond to physical locations in a location space. The system my segment the location space into a plurality of location bins each having one or more nodes. The system may generate two or more maps based on two or more mapping rule and map the first data set and the plurality of second data sets into the nodes of the location bins. The system may determine one or more target nodes in the maps that include the first data set and at least one of the second data sets. The system may determine a minimum distance between the first data set and one of the second data sets in a target node.

Merging data based on proximity and validation

A method is provided that includes assigning one or more first records of a first table and one or more second records of a second table to corresponding range groups of a plurality of range groups. The method further includes comparing at least one record of the first table with at least one record of the second table, where the at least one record of the first table and the at least one record of the second table are assigned to a first range group of the plurality of range groups. The method also includes, based on the comparing, identifying a parent record and a child record and segmenting the parent record based on the child record. Further, the method includes subsequent to the segmenting the parent record, performing a join operation on the first table and the second table.

Growing community of inventors

Gush-Dan, Israel

Avishay Meron