CN108737859A - Video recommendation method based on barrage and device - Google Patents

Abstract

The present invention discloses a video recommendation method and device based on bullet chatting. The method includes: according to the corpus of bullet chatting, model training is performed on word vectors and emotional labels of the corpus through a neural network, and the emotion of bullet chatting is predicted; The barrage published in the video builds the user into a user vector that can represent the user's preference for various videos, where each dimension of the user vector represents the user's preference for one category of various videos; through the PV-DM model Obtain the first vector of the text introduction of the video, and obtain the second vector of the distribution of the hidden topics of the bullet chat of the video through the topic generation model. The first vector and the second vector jointly construct the video vector; according to the user similarity and video similarity The degree predicts the user's prediction score for the video, and recommends the video with the prediction score not less than the predetermined threshold to the user. After calculating the scores for the similarity between users and the similarity between videos, the scores are weighted to obtain the prediction score.

Description

Barrage-based video recommendation method and device

technical field

The present invention relates to the technical field of video services, in particular to a barrage-based video recommendation method and device.

Background technique

With the development of online video sites, barrage has become a popular way of communication. There are a large number of videos in online video sites, and recommender systems have been used in video sites to improve user experience. At present, there are generally two recommendation methods: one is to match the title or introduction of the video according to the keywords entered by the user, and return videos with a higher degree of relevance; the other is to analyze the user's historical viewing records, ratings and comments and other data, analyze user preferences, and recommend videos that users may like.

The first method relies on the keywords entered by the user for matching. When the user cannot accurately describe his needs or the keywords entered are wrong, it cannot provide accurate recommendation results. The second method mainly relies on the user rating matrix and keywords extracted from user barrage for recommendation. However, in mainstream video websites, users cannot directly rate videos, so the user rating matrix cannot be obtained directly. However, the method of extracting keywords has the following problems: (1) the user's emotion to the keyword is not necessarily positive, if the user does not like the keyword, the video recommended to him cannot improve the user experience; (2) The keyword does not fully represent the interest of the user, and may be a word occasionally mentioned by the user, so the recommended video cannot improve the user experience.

Contents of the invention

In view of the above problems, the present invention establishes a user model and predicts the user's emotion by analyzing the user's barrage content, and recommends videos that the user may like.

In the first aspect, the embodiment of the present invention provides a bullet chat-based video recommendation method, including: according to the bullet chat corpus, perform model training on the word vector and emotion label of the corpus through a neural network, and predict the emotion of the bullet chat; Through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, where each dimension of the user vector represents the user's preference for a category of various videos; through PV -The DM model obtains the first vector of the text introduction of the video, and obtains the second vector of the distribution of each hidden topic of the video barrage through the topic generation model, and the first vector and the second vector jointly construct the video vector; according to the user similarity and video similarity to predict the user's predicted score for the video, and recommend the video with the predicted score not less than the predetermined threshold to the user, wherein the similarity between users and the similarity between videos are calculated after scoring, and the scores are weighted to get predictive score.

Optionally, use the Skip-gram model to train word vectors, and use the long-short-term memory network to predict the emotional polarity of the barrage sent by the user in the video.

Optionally, each dimension of the user vector includes a score obtained according to the emotion of the barrage, and the score is used to indicate the degree of liking.

Optionally, the similarity between users and the similarity between videos are respectively obtained by cosine similarity.

Optionally, in the recommending step, a plurality of tags that appear most often when the user watches videos within a predetermined time period are selected, and a set of videos to be predicted is obtained according to the tags.

In the second aspect, the present invention discloses a video recommendation device based on barrage, including: a sentiment analysis model training module, which is used to perform model training on word vectors and emotional tags of the corpus through a neural network based on the corpus of the barrage, And predict the emotion of described barrage; User model construction module, be used for by the described barrage that user releases in video, described user is constructed as the user vector that can represent described user to the degree of preference of various videos, Wherein each dimension of the user vector represents the degree of preference of the user to one class of videos; the video model building block is used to obtain the first vector of the text introduction of the video through the PV-DM model, Obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, the first vector and the second vector jointly construct the video vector; the recommendation module is used for user similarity and video The similarity predicts the user's predicted score for the video, and recommends the video with the predicted score not less than a predetermined threshold to the user, wherein after calculating the scores for the similarity between users and the similarity between videos, The scores are weighted to obtain the predicted score.

Optionally, use the Skip-gram model to train word vectors, and use the long-short-term memory network to predict the emotional polarity of the barrage sent by the user in the video.

Optionally, each dimension of the user vector includes a score obtained according to the emotion of the barrage, and the score is used to indicate the degree of liking.

Optionally, the similarity between users and the similarity between videos are respectively obtained by cosine similarity.

Optionally, in the recommending step, a plurality of tags that appear most often when the user watches videos within a predetermined time period are selected, and a set of videos to be predicted is obtained according to the tags.

In a third aspect, an embodiment of the present invention provides a computing device, including a memory, and one or more processors; wherein, the computing device further includes: one or more units, one or more units are stored in the memory and It is configured to be executed by one or more processors, and one or more units include instructions for performing the following steps: according to the corpus of bullet chatting, perform model training on word vectors and emotional labels of the corpus through a neural network, and predict bullet chat The emotion of the screen; through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, where each dimension of the user vector represents the user's preference for a category of various videos Degree; the first vector of the text introduction of the video is obtained by the PV-DM model, the second vector of the distribution of each hidden theme of the barrage of the video is obtained by the topic generation model, and the first vector and the second vector jointly construct the video vector; According to the user similarity and video similarity, predict the user's predicted score for the video, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein after calculating the scores for the similarity between users and the similarity between videos, The scores are weighted to obtain a predicted score.

Furthermore, an embodiment of the present invention provides a computer program product used in combination with a computing device, which is characterized by including a computer-readable storage medium and a computer program mechanism embedded therein; wherein the computer program mechanism includes instructions for performing the following steps : According to the corpus of bullet chatting, the neural network is used to perform model training on the word vectors and emotional labels of the corpus, and predict the emotion of the bullet chatting; through the bullet chatting posted by users in the video, the user is constructed to be able to express the user's interest in various videos The user vector of the degree of preference, where each dimension of the user vector represents the user’s preference for a category of videos; the first vector of the text introduction of the video is obtained through the PV-DM model, and the elasticity of the video is obtained through the topic generation model. The second vector of the distribution of each hidden theme of the episode, the first vector and the second vector jointly construct the video vector; predict the user's prediction score for the video according to the user similarity and video similarity, and the video with the prediction score not less than the predetermined threshold Recommend to the user, wherein after calculating the scores for the similarity between users and the similarity between videos, the scores are weighted to obtain the prediction score.

Compared with the prior art, the main difference and effect of the embodiment of the present invention is that the technical solution of the present invention can predict the user's real emotional expression according to the user's barrage, and recommend the user's favorite video.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings that need to be used in the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained based on these drawings without creative effort.

Fig. 1 is a flowchart of a bullet chatting-based recommendation method 100 according to a first embodiment of the present invention.

Fig. 2 is a schematic diagram of an LSTM model predicting emotional polarity of a barrage according to an embodiment of the present invention.

Fig. 3 is a structural block diagram of a barrage-based recommendation device 300 used in the third embodiment of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the implementation manner of the present invention will be further described in detail below in conjunction with the accompanying drawings. The illustrative system and method embodiments described herein are not intended to be limiting.

first embodiment

Fig. 1 is a flowchart of a bullet chatting-based recommendation method 100 according to an embodiment of the present invention. As shown in Figure 1, the specific processing flow of this method is as follows:

S110. According to the corpus of the bullet chatting, perform model training on the word vectors and emotion tags of the corpus through a neural network, and predict the emotion of the bullet chatting.

According to the embodiments of the present invention, it can be understood that it is necessary to collect various video-related data, such as video information, user information, user barrage, and emotional tags. Among them, video information includes but not limited to video number, video type label, video text introduction, etc.; user information includes but not limited to user number, etc.; user barrage can include user number, video number, barrage sending time, barrage playing time , barrage content, etc. Emotional tags are positive, negative and neutral tags that indicate emotional orientation for the user's barrage content.

Since the barrage is usually short and highly random, it is difficult for the existing word segmentation algorithm to achieve satisfactory results. Therefore, the embodiment of the present invention learns the vector representation of the words of the barrage, and uses it for sentiment analysis. As an example, an existing skip-gram model can be used to train word vectors.

According to an embodiment of the present invention, an LSTM (Long Short-Term Memory, long-short-term memory network) model is used to predict the emotional polarity of the barrage. As shown in Figure 2, the user’s barrage is extracted as a sequence of words, and the word vectors w _i (i=1, 2, ..., n) corresponding to the words are sequentially input into the LSTM unit, and each LSTM unit will output A h _i (i=1, 2, ..., n), the last hidden layer output vector h _{n is} used as a sentence representation and input to the softmax layer. The Softmax layer divides the vector into three labels: positive, negative and neutral. This method uses cross entropy as the loss function:

Among them, _DT represents all training barrages, and C is the number of emotional tags. The values of g _c (d) are 1 or 0, which respectively represent that the barrage d belongs to or does not belong to the current emotional label. P _c (d) represents the probability that the softmax layer classifies the barrage d into category c. It can be understood that model training uses backpropagation and stochastic gradient descent for parameter update.

S120. Using the barrage posted by the user in the video, construct the user as a user vector that can represent the user's preference for various videos, wherein each dimension of the user vector represents the user's preference for The liking degree of one category among the various types of videos.

In order to measure the similarity of users, the embodiment of the present invention calculates the similarity between users by calculating the user's preference for each video tag. Assuming that the video tag set of the entire video website is L, the feature of user i is expressed as |L|-dimensional vector u _i , and each dimension j (j=1, .., |L|) represents the user's preference for tag l _j degree. User vectors can be constructed and the similarity between users measured in the following ways:

(1) Obtain the video that the user has sent the barrage, denoted as V _i ;

(2) For each video v∈V _i , obtain its label set L _v , and update its user vector u _i for the barrage d sent by user i in this video:

u _i [L _v ]＝u _i [L _v ]+score _sentiment (d)

It means that the emotion of the bullet chatting d sent by the user in the video v is the user's emotion on the label of the video v, and the dimension corresponding to the label of the video v in the vector u _i will add the score obtained according to the emotion of the bullet chatting d score _sentiment (d). In the embodiment of the present invention, bullet screen emotion is divided into positive (positive), negative (negative) and neutral (neutral), thus score _sentiment (d) is defined as:

In this way, the user vector u _i represents user i's preference for various types of videos.

(3) After constructing the user vector, the cosine similarity can be used to measure the similarity between users:

S130. Obtain the first vector of the text introduction of the video through the PV-DM model, obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, the first vector and the The second vectors together construct the video vector.

Furthermore, the embodiment of the present invention vectorizes the video from two aspects: the text introduction of the video and the barrage of the video. As an example, collect all video introductions and video barrage, and use PV-DM (DistributedMemory Model of Paragraph Vectors, sentence vector distribution memory model) model training to get the vector of the text introduction part At the same time, the bullet chat of each video is used as a document, and the LDA (Latent Dirichlet Allocation) document topic generation model is used to obtain the distribution of the hidden topic of the bullet chat document. In this way, the _barrage of video vi can be expressed as a T-dimensional vector T is the number of hidden topics. The vector v _i of video i can be obtained by concatenating the two vectors after normalization.

It can be understood that, similar to measuring user similarity, cosine similarity can also be used to measure the similarity between videos:

Further, S140. Predict the user's predicted score for the video according to the user similarity and video similarity, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users is respectively calculated After the similarity score between the video and the video is calculated, the score is weighted to obtain the prediction score.

If the scores are predicted for all videos, the efficiency is relatively low. Therefore, in the embodiment of the present invention, a series of videos may be selected as candidates first. As an example, among the m videos watched by user i within a recent period of time, such as a few days or a week, collect the tags of these videos, select the tags with the most k occurrences, and find all the videos under these tags as a candidate set , denoted as V _can . According to an embodiment of the present invention, the user's score is predicted from two aspects of the similarity between users and the similarity between videos. For each video v∈V _can , user i’s possible scores are:

score(i, v)=k ₁ ·score _user +k ₂ ·score _video

Where score _user represents the predicted score obtained from user similarity, score _video represents the predicted score obtained from video similarity, where k ₁ and k ₂ are parameters, k ₁ +k ₂ =1, k ₁ and k ₂ represent focus on these two aspects. score _user and score _video can be expressed as:

where D _v represents the collection of bullet chatting in video v, |D _v | represents the size of the collection, and u _d represents the vector of users who sent bullet chatting d. L _v represents the label set of video v, V _l represents the video set with the same label l as video v, and |V _l | is the size of the set. s(u _d , u _i ) and s(v _j , v) denote the similarity between users and the similarity between videos.

And sp(v _j ) represents the overall evaluation of video v _j by users watching video v _j , which can be expressed as:

Therefore, the score of user i can be predicted, and the n videos with the highest predicted scores in _{Vcan can} be returned to user i.

Therefore, the bullet chatting-based recommendation method of the embodiment of the present invention can predict the user's real emotional expression according to the user's bullet chatting, and recommend videos that the user likes.

All method implementations of the present invention can be implemented in software, hardware, firmware and other ways. Regardless of whether the invention is implemented in software, hardware, or firmware, the instruction codes may be stored in any type of computer-accessible memory (e.g., permanent or modifiable, volatile or nonvolatile, solid-state or non-solid state, fixed or replaceable media, etc.). Similarly, the memory can be, for example, Programmable Array Logic (Programmable Array Logic, referred to as "PAL"), Random Access Memory (Random Access Memory, referred to as "RAM"), Programmable Read Only Memory (Programmable Read Only Memory, referred to as "PROM") ), Read-Only Memory (Read-Only Memory, referred to as "ROM"), Electrically Erasable Programmable Read-Only Memory (Electrically Erasable Programmable ROM, referred to as "EEPROM"), magnetic disk, optical disc, Digital Versatile Disc (Digital Versatile Disc, referred to as "DVD") and so on.

second embodiment

Fig. 3 is a schematic block diagram of a barrage-based recommendation device 300 according to an embodiment of the present invention. This device is used for carrying out above-mentioned method flow process, comprises:

The sentiment analysis model training module 310 is configured to perform model training on word vectors and emotion tags of the corpus through a neural network according to the corpus of the bullet chat, and predict the emotion of the bullet chat.

The user model construction module 320 is used to construct the user into user vectors that can represent the user's preference for various types of videos through the barrage posted by the user in the video, wherein each of the user vectors Dimension represents the degree of preference of the user to one category of the various types of videos.

The video model construction module 330 is used to obtain the first vector of the text introduction of the video through the PV-DM model, and obtain the second vector of the distribution of each hidden topic of the barrage of the video through the topic generation model, the said The first vector and the second vector jointly construct a video vector.

The recommendation module 340 is used to predict the user's predicted score for the video according to the user similarity and video similarity, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users is respectively After calculating the score of the similarity between the degree and the video, the score is weighted to obtain the prediction score.

Optionally, use the Skip-gram model to train word vectors, and use the long-short-term memory network to predict the emotional polarity of the barrage sent by the user in the video.

Optionally, each dimension of the user vector includes a score obtained according to the emotion of the barrage, and the score is used to indicate the degree of liking.

Optionally, the similarity between users and the similarity between videos are respectively obtained by cosine similarity.

Optionally, in the recommending step, a plurality of tags that appear most often when the user watches videos within a predetermined time period are selected, and a set of videos to be predicted is obtained according to the tags.

The first embodiment is a method implementation corresponding to this embodiment, and this embodiment can be implemented in cooperation with the first embodiment. The relevant technical details mentioned in the first embodiment are still valid in this embodiment, and will not be repeated here to reduce repetition. Correspondingly, the relevant technical details mentioned in this implementation manner can also be applied in the first implementation manner.

Therefore, the barrage-based recommendation device in the embodiment of the present invention can predict the user's real emotional expression according to the user's barrage, and recommend videos that the user likes.

Further, based on the same technical concept, embodiments of the present invention provide a computing device, including memory, and one or more processors; wherein, the computing device further includes: one or more units, and one or more units are Stored in memory and configured to be executed by one or more processors, the one or more units include instructions for:

According to the corpus of the bullet chatting, the neural network is used to train the model of the word vector and emotional label of the corpus, and predict the emotion of the bullet chatting;

Through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, where each dimension of the user vector represents the user's preference for one category of various videos;

Obtain the first vector of the text introduction of the video through the PV-DM model, obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, and the first vector and the second vector jointly construct the video vector;

According to the user similarity and video similarity, predict the user's predicted score for the video, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein after calculating the scores for the similarity between users and the similarity between videos, The scores are weighted to obtain a predicted score.

Based on the same technical idea, the embodiments of the present invention provide a computer program product used in combination with a computing device, which is characterized in that it includes a computer-readable storage medium and a computer program mechanism embedded therein; wherein, the computer program mechanism includes Instructions to perform the following steps:

According to the corpus of the bullet chatting, the neural network is used to train the model of the word vector and emotional label of the corpus, and predict the emotion of the bullet chatting;

Through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, where each dimension of the user vector represents the user's preference for one category of various videos;

Obtain the first vector of the text introduction of the video through the PV-DM model, obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, and the first vector and the second vector jointly construct the video vector;

According to the user similarity and video similarity, predict the user's predicted score for the video, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein after calculating the scores for the similarity between users and the similarity between videos, The scores are weighted to obtain a predicted score.

It should be appreciated that in the above description of exemplary embodiments of the invention, in order to streamline this disclosure and to facilitate understanding of one or more of the various inventive aspects, various features of the invention are sometimes grouped together in a single embodiment, figure, or in its description. This method of disclosure, however, is not to be interpreted as reflecting an intention that the claimed invention requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention.

Furthermore, those skilled in the art will understand that although some embodiments described herein include some features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention. and form different embodiments. For example, in the claims, any one of the claimed embodiments can be used in any combination.

Although various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are for purposes of illustration and not limitation, with the true scope to be indicated by the appended claims, along with the full scope of equivalents to which such claims are entitled. It is also to be understood that terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting.

Claims (12)

1. A barrage-based recommendation method, comprising: Sentiment analysis model training step, according to the corpus of barrage, carry out model training to the word vector of described corpus and emotion label by neural network, and predict the emotion of described barrage; The user model construction step is to construct the user as a user vector that can represent the user's preference for various types of videos through the barrage posted by the user in the video, wherein each dimension of the user vector represents the Describe the user's preference for one of the various types of videos; The video model construction step obtains the first vector of the text introduction of the video by the PV-DM model, obtains the second vector of the distribution of each hidden theme of the barrage of the video by the topic generation model, and the first vector Constructing a video vector jointly with the second vector; The recommendation step is to predict the user's predicted score for the video according to the user similarity and video similarity, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users and the video After the similarity scores are calculated, the scores are weighted to obtain the prediction score. 2. The recommendation method based on bullet chatting according to claim 1, characterized in that the word vector is trained using the Skip-gram model, and the long-short-term memory network is used to predict the value of the bullet chatting sent by the user in the video. emotional polarity. 3. The recommendation method based on bullet chatting according to claim 1, wherein each dimension of the user vector includes a score obtained according to the emotion of the bullet chatting, and the score is used to represent the degree of preference . 4. The barrage-based recommendation method according to claim 1, wherein the similarity between the users and the similarity between the videos are respectively obtained by cosine similarity. 5. The barrage-based recommendation method according to claim 1, wherein, in the recommending step, select a plurality of tags that appear most frequently when the user watches a video within a predetermined time, and obtain the required tags according to the tags. Predicted collection of videos. 6. A barrage-based recommendation device, characterized by comprising, Sentiment analysis model training module, for according to the corpus of barrage, carry out model training to the word vector of described corpus and emotion label by neural network, and predict the emotion of described barrage; The user model construction module is used to construct the user into a user vector that can represent the user's preference for various types of videos through the barrage posted by the user in the video, wherein each dimension of the user vector Indicates the degree of preference of the user to one category of the various types of videos; The video model construction module is used to obtain the first vector of the text introduction of the video by the PV-DM model, and obtain the second vector of the distribution of each hidden theme of the barrage of the video by the topic generation model, the second vector A vector and the second vector jointly construct a video vector; The recommendation module is used to predict the user's predicted score for the video according to the user similarity and video similarity, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users is calculated separately After the similarity score between the video and the video is calculated, the score is weighted to obtain the prediction score. 7. The recommendation device based on bullet chatting according to claim 6, wherein the word vector is trained using a Skip-gram model, and a long-short-term memory network is used to predict the value of the bullet chatting sent by the user in the video. emotional polarity. 8. The recommendation device based on bullet chatting according to claim 6, wherein each dimension of the user vector includes a score obtained according to the emotion of the bullet chatting, and the score is used to represent the degree of preference . 9. The barrage-based recommendation device according to claim 6, wherein the similarity between the users and the similarity between the videos are respectively obtained by cosine similarity. 10. The barrage-based recommendation device according to claim 6, characterized in that, in the recommendation module, select a plurality of tags that appear most frequently when the user watches a video within a predetermined time, and obtain the required tags according to the tags. Predicted collection of videos. 11. A computing device comprising a memory, and one or more processors; wherein the computing device further comprises: one or more units stored in the memory and configured to be executed by the one or more processors, the one or more units comprising instructions for performing : According to the corpus of the barrage, carry out model training on the word vector and the emotion label of the corpus through a neural network, and predict the emotion of the barrage; Through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, wherein each dimension of the user vector represents the user's preference for the The preference level of one category in various videos; Obtain the first vector of the text introduction of the video through the PV-DM model, obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, the first vector and the second vectors together to construct video vectors; According to the user similarity and video similarity, predict the user's predicted score for the video, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users and the video After the similarity score is calculated, the score is weighted to obtain the prediction score. 12. A computer program product for use in combination with the computing device of claim 11 , comprising a computer-readable storage medium and a computer program mechanism embedded therein; wherein the computer program mechanism includes Instructions to perform the following steps: According to the corpus of the barrage, carry out model training on the word vector and the emotion label of the corpus through a neural network, and predict the emotion of the barrage; Through the barrage posted by the user in the video, the user is constructed as a user vector that can represent the user's preference for various videos, wherein each dimension of the user vector represents the user's preference for the The preference level of one category in various videos; Obtain the first vector of the text introduction of the video through the PV-DM model, obtain the second vector of the distribution of each hidden theme of the barrage of the video through the topic generation model, the first vector and the second vectors together to construct video vectors; According to the user similarity and video similarity, predict the user's predicted score for the video, and recommend the video with the predicted score not less than a predetermined threshold to the user, wherein the similarity between users and the video After the similarity score is calculated, the score is weighted to obtain the prediction score.