# US 10,311,361 B1 · Media theme propagation using decretization of continuous variables > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,311,361 B1 - Document ID: US10311361B1 - Status: Granted - Kind: B1 - Application number: 14/315,167 - Filed: 2014-06-25 - Granted: 2019-06-04 - Earliest claimed: 2014-06-25 - Assignees: Amazon Technologies, Inc. - Applicants: Amazon Technologies, Inc. - Inventors: Brandon Scott Durham; Haowei Lu; Christopher Lon McGilliard; Darren Levi Malek; Joshua Fredrick Lutes; Toby Ray Latin-Stoermer - Web page: https://patentgrep.com/patent/US10311361B1 ## CPC - G06N 5/04: Inference or reasoning models - G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks - G06N 20/00: Machine learning - H04N 21/4668: {for recommending content, e.g. movies} - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - H04N: PICTORIAL COMMUNICATION, e.g. TELEVISION ## Abstract A technology for propagating themes is provided. In one example, a method may include identifying media having a content feature contained in a presentation of the media. The method may include extracting the content feature of the media as a continuous variable and discretizing the continuous variable into a bucket representing a discrete value. A theme label may be applied to the media and may be propagated to other media with continuous variables discretized into the subset of buckets. ## Claims ### Claim 1 A computer-implemented method for propagating themes, comprising: identifying a digital music segment in a media data store having an audio feature; extracting, by a computing system, the audio feature of the digital music segment as a continuous variable normalized within a range from 0 to 1 and having a scaling factor which varies between audio features, and a magnitude of the continuous variable within the range being proportional to a contribution of the continuous variable to a user-selected theme label to be applied to the digital music; discretizing in a MapReduce job, by the computing system, the continuous variable extracted from the digital music segment into a subset of discrete value buckets representing the continuous variable; applying the user-selected theme label to the digital music; and propagating the theme label to other digital music with continuous variables discretized into the subset of buckets using an Adsorption-based theme propagation model; validating that the user-selected theme label propagated to other digital music is associated with a pre-specified label using a MapReduce runner; generating a learning dataset having a defined set of themes based on the validating; and categorizing unlabeled digital media in a digital media catalog to the set of themes using the learning dataset. ### Claim 2 (depends on 1) The method of claim 1, further comprising using a bipartite graph to map known theme labels from training data to the digital music segment using the audio feature and Adsorption machine learning. ### Claim 3 (depends on 1) The method of claim 1, wherein the audio feature is extracted from a timbre of the digital music. ### Claim 4 A computing device that is configured to facilitate propagation of themes, comprising: a processor; a memory in electronic communication with the processor; instructions stored in the memory, the instructions being executable by the processor to perform a method, comprising: identifying media in a media data store having a content feature contained in a presentation of the media; extracting the content feature of the media as a continuous variable normalized within a range from 0 to 1, a magnitude of the continuous variable within the range being proportional to a contribution of the continuous variable to a theme label to be applied to the media; discretizing the continuous variable into a bucket representing a discrete value, using a MapReduce job; applying the theme label to the media; and propagating the theme label to other media with continuous variables discretized into the bucket; validating that the user-selected theme label propagated to other digital music is associated with a pre-specified label using a MapReduce runner; generating a learning dataset having a defined set of themes based on the validating; and categorizing unlabeled digital media in a digital media catalog to the set of themes using the learning dataset. ### Claim 5 (depends on 4) The method of claim 4, further comprising identifying the theme label to apply to the media from training media having the content feature. ### Claim 6 (depends on 4) The method of claim 4, further comprising generating a bipartite graph linking the content feature and the bucket for the media, the bucket representing a discretized continuous variable. ### Claim 7 (depends on 4) The method of claim 4, further comprising discretizing the continuous variable into a plurality of buckets when the continuous variable is within a predetermined boundary proximity threshold to a boundary between the plurality of buckets, and discretizing the continuous variable into a single bucket when the continuous variable is outside the predetermined boundary proximity threshold. ### Claim 8 (depends on 7) The method of claim 7, wherein the plurality of buckets include buckets representing the continuous variable. ### Claim 9 (depends on 4) The method of claim 4, further comprising analyzing a waveform of the media to extract the content feature as the continuous variable. ### Claim 10 (depends on 9) The method of claim 9, wherein the media comprises digital music. ### Claim 11 (depends on 4) The method of claim 4, further comprising forming a bipartite graph including the media and training media. ### Claim 12 (depends on 4) The method of claim 4, wherein the media is digital music. ### Claim 13 (depends on 4) The method of claim 4, wherein the media comprises at least one of audio, video or text media. ### Claim 14 (depends on 4) The method of claim 4, further comprising performing Adsorption to propagate the theme label from the bucket. ### Claim 15 A non-transitory computer-readable medium comprising computer-executable instructions which, when executed by a processor, implement a system, comprising: a training data store storing a training dataset having a defined set of themes to apply theme labels to unlabeled music; a media data store storing the unlabeled music; a feature extractor configured to extract a feature from the unlabeled music as a continuous variable normalized within a range from 0 to 1 and having a scaling factor which varies between audio features; a feature discretizer in a MapReduce job configured to discretize the continuous variable into a bucket as a discrete value; and a label generator configured to use Adsorption machine learning to generate and propagate theme labels for the unlabeled music based on the discrete value in the bucket for the unlabeled music; validate that the user-selected theme label propagated to other digital music is associated with a pre-specified label using a MapReduce runner; and categorize the unlabeled music in a digital media catalog to the set of themes based on the validating. ### Claim 16 (depends on 15) The system of claim 15, further comprising a graph generator to generate a bipartite graph for the training dataset and the unlabeled music. ### Claim 17 (depends on 15) The system of claim 15, further comprising an Adsorption propagation engine to propagate the theme labels to additional unlabeled music using the feature for the unlabeled music.