Qi Guo

20200401594, Dec 24, 2020

Jun 21, 2019

16/449122

- Redmond WA, US
Dan Liu - Santa Clara CA, US
Qi Guo - Sunnyvale CA, US

G06F 16/2457
G06F 17/18
G06F 17/15
G06N 3/04

In an example embodiment, a platform is provided that utilizes information available to a computer system to feed a neural network. The neural network is trained to determine both the probability that a searcher would select a given potential search result if it was presented to him or her and the probability that a subject of the potential search result would respond to a communication from the searcher. These probabilities are combined to produce a single score that can be used to determine whether to present the searcher with the potential search result and, if so, how high to rank the potential search result among other search results. During the training process, a rescaling transformation for each input feature is learned and applied to the values for the input features.

20200401627, Dec 24, 2020

Jun 21, 2019

16/449110

- Redmond MA, US
Daniel Sairom Krishnan Hewlett - Sunnyvale CA, US
Qi Guo - Sunnyvale CA, US
Wei Lu - Sunnyvale CA, US
Xuhong Zhang - Sunnyvale CA, US
Wensheng Sun - Sunnyvale CA, US
Mingzhou Zhou - Sunnyvale CA, US
Anthony Hsu - Sunnyvale CA, US
Keqiu Hu - Sunnyvale CA, US
Yi Wu - Sunnyvale CA, US
Chenya Zhang - Santa Clara CA, US
Baolei Li - Santa Clara CA, US

G06F 16/9038
G06N 3/02

In an example embodiment, a platform is provided that utilizes information available to a computer system to feed a neural network. The neural network is trained to determine both the probability that a searcher would select a given potential search result if it was presented to him or her and the probability that a subject of the potential search result would respond to a communication from the searcher. These probabilities are essentially combined to produce a single score that can be used to determine whether to present the searcher with the potential search result and, if so, how high to rank the potential search result among other search results. In a further example embodiment, embeddings used for the input features are modified during training to maximize an objective.

20200401643, Dec 24, 2020

Jun 21, 2019

16/449135

- Redmond WA, US
Daniel Sairom Krishnan Hewlett - Sunnyvale CA, US
Qi Guo - Sunnyvale CA, US

G06F 16/9538
G06N 3/04
G06F 16/906

In an example embodiment, position bias is addressed by introducing an inverse propensity weight into a loss function used to train a machine-learned model. This inverse propensity weight essentially increases the weight of candidates in the training data that were presented lower in a list of candidates. This achieves the benefit of counteracting the position bias and increases the effectiveness of the machine-learned model in generating scores for future candidates. In a further example embodiment, a function is generated for the inverse propensity weight based on responses to contact requests from recruiters. In other words, while the machine learned-model may factor in both the likelihood that a recruiter will want to contact a candidate and the likelihood that a candidate will respond to such a contact, the function generated for the inverse propensity weight will be based only on training data where the candidate actually responded to a contact.

20200401644, Dec 24, 2020

Jun 21, 2019

16/449149

- Redmond WA, US
Dan Liu - Santa Clara CA, US
Qi Guo - Sunnyvale CA, US
Wenxiang Chen - Sunnyvale CA, US
Xiaoyi Zhang - Sunnyvale CA, US
Lester Gilbert Cottle - Sunnyvale CA, US
Xuebin Yan - Sunnyvale CA, US
Yu Gong - Santa Clara CA, US
Haitong Tian - San Jose CA, US
Siyao Sun - Mountain View CA, US
Pei-Lun Liao - Sunnyvale CA, US

G06F 16/9538
G06N 20/00
G06F 17/27
G06N 3/04

In an example embodiment, position bias and other types of bias may be compensated for by using two-phase training of a machine-learned model. In a first phase, the machine-learned model is trained using non-randomized training data. Since certain types of machine-learned models, such as those involving deep learning (e.g., neural networks) require a lot of training data, this allows the bulk of the training to be devoted to training using non-randomized training data. However, since this non-randomized training data may be biased, a second training phase is then used to revise the machine-learned model based on randomized training data to remove the bias from the machine-learned model. Since this randomized training data may be less plentiful, this allows the deep learning machine-learned model to be trained to operate in an unbiased manner without the need to generate additional randomized training data.

20200004835, Jan 2, 2020

Jun 28, 2018

16/021667

- Redmond WA, US
Gungor Polatkan - San Jose CA, US
Qi Guo - Sunnyvale CA, US
Krishnaram Kenthapadi - Sunnyvale CA, US
Sahin Cem Geyik - Redwood City CA, US

G06F 17/30
G06N 3/02

Techniques for generating candidates for search using a scoring and retrieval architecture and deep semantic features are disclosed herein. In some embodiments, a computer system generates a profile vector representation for user profiles based profile data, stores the profile vector representations, receives a query subsequent to the storing of the profile vector representations, generates a query vector representation for the query, retrieves the stored profile vector representations of the user profiles based on the receiving of the query, generates a corresponding score for pairings of the user profiles and the query based on a determined level of similarity between the profile vector representation of the user profiles and the query vector representation, and causes an indication of at least a portion of the user profiles to be displayed as search results for the query based on the generated scores of the user profiles.

20200004886, Jan 2, 2020

Jun 28, 2018

16/021639

- Redmond WA, US
Gungor Polatkan - San Jose CA, US
Qi Guo - Sunnyvale CA, US
Krishnaram Kenthapadi - Sunnyvale CA, US
Sahin Cem Geyik - Redwood City CA, US

G06F 17/30
G06N 99/00

Techniques for generating supervised embedding representations for search are disclosed herein. In some embodiments, a computer system receives training data comprising query representations including an entity included in a corresponding search query submitted by a querying user, search result representations for each one of the query representations, and user actions for each one of the query representations, and generates a corresponding embedding vector for each one of the at least one entity using a supervised learning algorithm and the received training data. In some example embodiments, the corresponding search result representations for each one of the query representations represents a plurality of candidate users displayed in response to the search queries based on profile data of the candidate users, and the user actions comprise actions by the querying user directed towards at least one candidate user in the search results.

20200005134, Jan 2, 2020

Jun 28, 2018

16/021654

- Redmond WA, US
Gungor Polatkan - San Jose CA, US
Qi Guo - Sunnyvale CA, US
Krishnaram Kenthapadi - Sunnyvale CA, US
Sahin Cem Geyik - Redwood City CA, US

G06N 3/08
G06F 17/30
G06N 3/04

Techniques for generating supervised embedding representations using unsupervised embedding representations and deep semantic structured models for search are disclosed herein. In some embodiments, a computer system generates a graph data structure based on accessed profile data, generates an initial embedding vector using an unsupervised machine learning algorithm, receiving training data comprising query representations, search result representations, and user actions, with each one of the plurality of query representations comprising the initial embedding vector, and generates a final embedding vector using a supervised learning algorithm and the received training data.

20200005149, Jan 2, 2020

Jun 28, 2018

16/021692

- Redmond WA, US
Gungor Polatkan - San Jose CA, US
Qi Guo - Sunnyvale CA, US
Krishnaram Kenthapadi - Sunnyvale CA, US
Sahin Cem Geyik - Redwood City CA, US

Microsoft Technology Licensing, LLC - Redmond WA

G06N 3/08
G06F 17/30
G06N 3/04

Techniques for applying learning-to-rank with deep learning models for search are disclosed herein. In some embodiments, a computer system trains a ranking model using training data and a loss function, with the ranking model comprising a deep learning model and being configured to generate similarity scores based on a determined level of similarity between profile data of reference candidates users in the training data and reference query data of reference queries in the training data. The computer system receives a target query comprising target query data from a computing device of a target querying user, and then generates a corresponding score for target candidate users based on a determined level of similarity between profile data of the target candidate users and the target query data using the trained ranking model.