Combined wide and deep machine learning models for automated database element processing systems, methods and apparatuses
A method of automated database element processing includes training a wide machine learning model with historical feature vector inputs to generate a wide ranked element output. The method includes training a deep machine learning model with the historical feature vector inputs to generate a deep ranked element output. The method includes generating a set of inputs specific to an individual entity, obtaining a set of current article database elements, and creating a feature vector input according to the set of inputs and the set of current article database elements. The method includes processing the feature vector input with the wide machine learning model to generate a wide ranked element list, processing the feature vector input with the deep machine learning model to generate a deep ranked element list, and merging database elements of the wide and deep ranked element lists to generate a ranked element recommendation output.
1 . A computer-implemented method of managing news recommendation filter bubbles, comprising:
receiving, via at least one processor, a set of articles previously accessed by a user;
determining whether a number of articles previously accessed by the user satisfies read articles count criteria;
identifying, via the at least one processor, patterns from the set of articles previously accessed by the user;
identifying, via the at least one processor, terms opposing the patterns in the set of articles previously accessed by the user;
obtaining, via the at least one processor, a set of candidate articles having opposing content to the patterns based on the identified opposing terms;
receiving, via the at least one processor, user configuration input specifying how opposing content articles should be presented;
generating, via the at least one processor, a recommendation list including at least one article having opposing content according to the user configuration input; and
displaying on a display device, via the at least one processor, the recommendation list to the user.
2 . The method of claim 1 , wherein the opposing content comprises factual articles.
3 . The method of claim 1 , wherein the opposing content comprises opinion articles.
4 . The method of claim 1 , further comprising quantifying semantics for obtaining the set of candidate articles, which includes:
training a wide machine learning model using unsupervised learning; and
training a deep machine learning model using unsupervised learning.
5 . The method of claim 4 , wherein the wide machine learning model comprises a term frequency-inverse document frequency (TFIDF) model.
6 . The method of claim 4 , wherein the deep machine learning model comprises a Doc2Vec model.
7 . The method of claim 1 , further comprising determining a freshness score for each article having opposing content.
8 . The method of claim 7 , wherein the freshness score is based on a time period between when a recommendation is generated and when the article was published.
9 . The method of claim 1 , further comprising determining a section score for each article having opposing content based on a section overlap with the set of articles previously accessed by the user.
10 . The method of claim 1 , wherein receiving user configuration input comprises receiving a selection between factual opposing content articles and opinion opposing content articles.
11 . The method of claim 1 , further comprising obtaining a popular article list when the number does not satisfy the read articles count criteria.
12 . The method of claim 1 , further comprising retraining a Word2Vec model periodically, thereby updating the identified opposing terms.
13 . The method of claim 1 , wherein analyzing the set of candidate articles comprises:
generating a feature vector for each candidate article; and
comparing the feature vector against the identified opposing terms.
14 . The method of claim 1 , further comprising calculating a similarity score between each candidate article and the articles previously accessed by the user.
15 . The method of claim 14 , further comprising removing candidate articles having a similarity score above a threshold to ensure opposing content.
16 . The method of claim 1 , wherein the recommendation list includes both similar content articles and opposing content articles.
17 . The method of claim 1 , further comprising identifying user engagement with recommended opposing content articles including at least one of the following: a number of user views, a view time, and an eye movement.
18 . The method of claim 1 , further comprising identifying the opposing terms via Word2Vec.
19 . The method of claim 1 , further comprising capturing a change of interest of the user based on at least one sliding window.
20 . The method of claim 1 , wherein generating the recommendation list includes processing the set of candidate articles by a trained wide machine learning model to generate a wide ranked element list, processing the set of candidate articles by a trained deep machine learning model to generate a deep ranked element list, and merging elements of the wide ranked element list and the deep ranked element list to generate the recommendation list.
21 . A computer-implemented method of managing news recommendation filter bubbles, comprising:
receiving, via at least one processor, a set of articles previously accessed by a user;
identifying, via the at least one processor, patterns from the set of articles previously accessed by the user;
identifying, via the at least one processor, terms opposing the patterns in the set of articles previously accessed by the user;
obtaining, via the at least one processor, a set of candidate articles having opposing content to the patterns based on the identified opposing terms;
calculating a similarity score between each candidate article and the articles previously accessed by the user;
removing candidate articles having a similarity score above a threshold to ensure opposing content;
receiving, via the at least one processor, user configuration input specifying how opposing content articles should be presented;
generating, via the at least one processor, a recommendation list including at least one article having opposing content according to the user configuration input; and
displaying on a display device, via the at least one processor, the recommendation list to the user.