Prediction of media success from plot summaries using machine learning model
Disclosed is a machine learning-based media success prediction through plot summaries According to an embodiment, a method comprises performing preprocessing on text data including a plot summary, calculating a sentiment score from the preprocessed text data using a first model, generating first input data using the calculated sentiment score, generating second input data from the preprocessed data using a second model, and determining a candidate class of content corresponding to the plot summary by applying the first input data and the second input data to a pre-trained third model. The candidate class includes a first class indicating success and a second class indicating failure.
1. A processor-implemented method, comprising:
performing preprocessing on text data and a success score, wherein the text data only includes a plot summary about a story of a content, and the success score corresponds to the content;
calculating a sentiment score for each of a plurality of sentences constituting the plot summary from the preprocessed text data using a first model, wherein the sentiment score which is an N-dimensional vector is calculated in reverse order from a last sentence among the plurality of sentences, and when a number of the sentences M is less than N, zero-padding is applied to as many remaining dimensions as N−M;
generating first input data as an output of a merged one-dimensional convolution neural network (1D CNN) or residual bidirectional long short-term memory (LSTM) by applying the calculated sentiment score to the merged 1D CNN or residual bidirectional LSTM;
generating second input data from the preprocessed text data using a second model including an embeddings from language models (ELMO) model; and
determining a candidate class of content corresponding to the plot summary by combining the first input data as a first output from the merged 1D CNN or residual bidirectional LSTM and the second input data as a second output from the ELMO embedding layer, and applying the first input data and the second input data to a pre-trained third model,
wherein the third model is a classification model pre-trained based on only a training plot summary about a story of a content, labeled with the success score corresponding to the content, and the training plot summary is the preprocessed text data,
wherein the candidate class includes a first class indicating success and a second class indicating failure, and
wherein the success score is an evaluation score previously evaluated by a consumer for the content.
2. The processor-implemented method of claim 1 , wherein the plot summary includes at least one of a movie, a musical, a concert, a play, a sports game, an exhibition, a book or music.
3. The processor-implemented method of claim 1 , wherein performing the preprocessing includes dividing the text data into sentences.
4. The processor-implemented method of claim 3 , wherein performing the preprocessing includes generating a list of the sentences of the text data.
5. The processor-implemented method of claim 1 , wherein the sentiment score includes a positive score, a negative score, a neutral score, or a compound score.
6. The processor-implemented method of claim 1 , wherein the first model is a neural network model trained by providing information for the sentiment score and the preprocessed text data, as training data.
7. The process-implemented method of claim 1 , wherein the first model is a valence aware dictionary and sentiment reasoner (VADER) sentiment analyzer.
8. The processor-implemented method of claim 1 , wherein generating the first input data includes generating a first feature vector by applying the sentiment score to the merged one-dimensional convolutional neural network (1D CNN).
9. The processor-implemented method of claim 1 , wherein:
generating the first input data includes:
generating a first vector by applying the sentiment score to a first bidirectional long short term memory (LSTM),
generating a second vector by applying the sentiment score to a second bidirectional LSTM; and
generating a second feature vector by adding the first vector and the second vector.
10. The processor-implemented method of claim 1 , wherein:
determining the candidate class includes:
generating a concatenated vector by concatenating the first input data and the second input data; and
determining the candidate class of the content corresponding to the plot summary by applying the concatenated vector to the pre-trained third model.
11. The processor-implemented method of claim 1 ,
wherein the evaluation score being X or more is classified as the first class, and the evaluation score being less than Y is classified as the second class.
12. The processor-implemented method of claim 11 , wherein X is different from Y.
13. A non-transitory computer system-readable recording medium recording a program for executing the method of claim 1 , on a computer system.