This work includes study and analysis of the functional implementation and usefulness of the Naive Bayesian classifier, especially working with medical data. This article presents a model for the classification of controlled moods based on a naive Bayes algorithm. Naive Bayes is known to be one of the simplest probability classifiers. Typically, it works extremely well under favorable circumstances, despite the fact that all functions are conditionally independent of a specific class. In order to train such a classifier, it is important to measure the probabilities of classes as well as their conditional probabilities, which will later be used for new classifications. Vector Machines Naive Bayesian Classifier, Sentiment Analysis, KNN-k-nearest neighbor algorithm, Support algorithm) and SVM(Support Vector Machines) experiments.
Work
Emotion recognition, in other words, the study of thoughts, is a large space for the study of
judgments, beliefs, behaviors, as objects of the emotional fund for something particular. An entity,
individual, product, or service, for example. At this stage, all of the above theoretical studies are
under the aegis of mood analysis and thought extraction. If we single out the industry, this word can
be found in a more scientific hue. The very analysis of sentence terms first appears in [
SA 's application is to collect input from consumers on the introduction of new goods, political
campaigns and even widespread in financial markets. The purpose of this strategy is to decide the
attitude of the narrator to any subject or simply to the contextual polarity of the paper. Early work in
this area was done by Terny and Peng ([
These days present day clinics are well prepared with observing and other information collection gadgets coming about in colossal information that are collected persistently through wellbeing examination and therapeutic treatment. All this is driven to the reality that the restorative zone produces progressively voluminous sums of electronic information which are getting to be more complicated.
Mood analysis is a challenging task, with the use of NB (Naive Bayes), K-NN (k-nearest neighbor
The area of big data and machine learning may be the functional field of application of the findings of scientific work.
2020 Copyright for this paper by its authors.
The goal of the work is to carry out a thorough analysis of the Naive Bayesian Classifier in comparison with some of the most common rivals of this technology in order to improve data processing. The proposed classification of the text, based on the collection of features and preprocessing, is therefore intended to serve as an opportunity to enhance the accuracy of the classification.
The key tasks in the course of the work are to establish the a priori concepts of the work of the Naive Bayesian classifier: describe the key characteristics, advantages and disadvantages of using NBC (Naive Bayesian Classifier) for sentimental analysis; define the key properties, advantages and disadvantages of using sentimental research help vector machines; define the key properties, advantages and disadvantages of using KNN-Method K of the closest neighbors for sentimental analysis;
The goal of the research is the problem of step-by-step data processing and the classification by
vector of sentimental analysis of all the above methods and the analysis of the consequent
optimization of its function. The topic of the research is the Naive Bayesian Classifier algorithm and
its efficiency, which is compared with competitive means [
Acuteness of the study: analysis of attitudes is a method of collecting knowledge from the perceptions of users. People's decisions are affected by the views of others. Today, if someone wants to buy a product or wants to watch a movie, he/she will first look for feedback and opinions about that product or movie on social networks, blogs, etc. When there is a massive influx of user opinions on social networks such as Twitter, Facebook and other user forums, it becomes very difficult to classify moods with this large data manually. There is also a need for an integrated mood analysis framework.
In the job, various testing methods are used. Theoretical research methods include: algorithm analysis, comparison, convergence approach from abstract to concrete. Empirical approaches, including comparison and calculation, are directly present
The NBC is based on the Bayesian law, with a clear presumption of freedom. The naive Bayesian
model presupposes a simplification of the conditional assumption of independence. In other words, a
class (positive or negative) is given whose words are conditionally independent of each other. This
assumption does not have a direct impact on the accuracy of the text classification, but actually allows
the quick classification algorithms applicable to this mission. In their 2003 paper, Rennes et al.
address the implementation of the Naïve Bayesian tasks of text classification. [
The main reason is that NB (Naïve Bayesian) with sampling tends to achieve a lower classification
error than the original [
2.1.
The NBC is based on the Bayesian law, with a clear presumption of freedom. The naive Bayesian
model presupposes a simplification of the conditional assumption of independence. In other words, a
class (positive or negative) is given whose words are conditionally independent of each other. This
assumption does not have a direct impact on the accuracy of the text classification, but actually allows
the quick classification algorithms applicable to this mission. In their 2003 paper, Rennes et al.
address the implementation of the Naïve Bayesian tasks of text classification. [
NBC is a tool that applies to a particular class of tasks, namely those that are formulated to connect
an object with a discreet category. From a community of numerical methods, the naive Bayes has a
range of advantages, such as simplicity, speed and high precision. K. Ming Leung [
The k-nearest neighbor algorithm (k-NN) is a method of classifying an object based on the
majority class among its nearest neighbors. KNN is a form of lazy learning in which the function is
only approximated locally and all calculations are deferred to classification. The KNN algorithm is
typically based on the Euclidean or Manhattan distance. However, you can use some other distance,
such as the Chebyshev standard or the Mahalanobis distance. The major downside of KNN is that it
uses all the functions to measure distance and costs a lot of time to identify objects [
SVM works well for text classification because of its advantages, such as its ability for processing
large items. Another benefit is that SVM is efficient when there are few instances, and also because
most of the problems are linearly separated. The reference vector machine has shown promising
results in previous studies in the field of mood analysis. [
The accuracy of the results of the intellectual study is directly influenced by the quality of the data. The pre-processing step is therefore necessary in order to achieve a better classification result and even to improve the time used to train and generalize the model.
3.1.
The data comes from Kaggle's call - "Bag of Words Meets Bags of Popcorn". There are 25.000 IMDB movie reviews that are either positive or negative. IMDB scores are considered to range from 0 to 10. The additional pre-processing step performed by the data set authors transforms the rating into binary moods. Of course, one film can have several ratings, but with a condition of no more than 30.
The id column combines the movie ID with a unique number of reviews.
We first use BeautifulSoup [
Here is what the same text looks like in its purified form in the figure
In this stage, the previous data is cleared, allowing you to continue the process and bring the data
to the state of the Words Bag model [
Naive Bayesian models are probabilistic classifiers used by the well-known Bayesian theorem, perform and make clear assumptions about the independence of data features.
Intuitively, this may sound like a crazy idea. The following statement is a well-known fact: the previous word has an effect on the present and the next. The belief, however, simplifies mathematics and works very well in reality.
( , ) ( ) ( ) ( ) (1) Where: specific class; a document that will be classified; a priori probability; a posteriori probability.
This equation gives us a conditional probability that event A will occur when B occurs. To find out, we need to measure the likelihood that B will happen if A occurs, and multiply it by the likelihood that A (known as the previous one) will happen. All of this is separated by the probability that B will happen on its own.
The naive assumption helps one to reformulate Bayes ' theorem as follows:
i (2) P(Sentiment ,..., n ) 1
P(1 ,...., n )
We just don't care about the odds. In a given case, we would like to know if the text has a positive or negative attitude. We can skip the denominator entirely, simply because it scales the numerator: (3) (Sentiment ,..., n ) P(Sentiment)in 1 P( Sentiment)
1 i
Thus, before choosing a feeling, we compare the scores for each feeling and select the one that has a higher score.
So, we're classifying the text into one of two groups / categories-positive and negative.
Multidimensional Naive Bayes helps us to present the features of the model in the form of the frequency of their occurrence (how much a term is present in our review). In other words, it informs us that the distributions of chance that we use are multinomial [13, 16].
The intuition of the classifier the text document is presented as if it were a bag of words, i.e. a split collection of words indicating their location with a bag of words, holding only their frequency in the document. In the example in the illustration, instead of reflecting the order of words in all phrases like "I love this movie" and "I would recommend it," we simply note that the word was repeated five times in the first passage, the word six times, the word love, recommend, film once, and so on.
The value of the α class can be positive / negative. A text is a summary of a particular film. The
Naive Bayes Multinomial Model [
Let us remember, first of all, the estimation of the highest probability. We're only going to use frequencies in the results. In the previous paper, we ask what percentage of the documents in our study set are in each class. Let the number of documents in our class C training details, and the total number of documents. And then: (4) ( )
N c
N
Where: the number of documents in the class; total number of documents.
Multinomial Naive Bayes helps you to present the characteristics of the models as the frequency of their occurrence.
The model is based on the multiplication of a number of probabilities. They can be so close that they're rounded to zero by the machine. We can therefore use the logarithmic probability: (5) log P(Sentiment ,..., n ) log P(Sentiment) log in 1 P( Sentiment)
1 i
There is, however, an issue with learning with the highest probability. It suffices to say that we are attempting to make a positive evaluation of the likelihood of the word "fantastic" in this class, but suppose that there are no educational documents that simultaneously contain the word "fantastic" and are graded as positive. Perhaps the word "fantastic" appears by chance (in a sarcastic/ironical sense) in the negative class. In this case, the likelihood for this function is zero: count(" fantastic", positive) (6) V count ( , positive) 0 Pˆ ( i c) count ( i , c) 1 (count ( , c) 1) ( count ( i , c) 1 count( , c) V
But since naive Bayes naively multiplies all the probabilities of traits together, zero probability in terms of plausibility for any class will cause the probability of that class to be zero, regardless of other proof.
V V
The problem with the MLE score is that it is zero for a term-class combination not contained in the training results. Training data are never large enough to accurately reflect the frequency of unusual occurrences. To remove the zero likelihood problem [12, 16], add-one or Laplace smoothing is used. This mainly adds one to each account. Add-one smoothing can be interpreted as a previous homogeneous one (each term occurs once for each class), which is then updated as the learning data is received. As a consequence, the probability of a document given by its class is the normal multinomial distribution previously presented in equation 2. Calculate the a priori probability of a positive negation using equation 5 as follows: 4
Let's calculate the maximum probabilistic smoothing of the Naive Bayesian estimate using equation 5: ( pos) (neg ) 3
(7) 4 Story is good but ending is so boring and sadly. neg
Test
5 I like director’s direction. The location place in movie is so
boring. But story is good.
x (neg doc5) 43 43 43 ( pos doc5) (neg doc5) ( pos doc5) – the maximum average probability of positive words in document 5 is maximum, so document 5 is positive.
We have an average classification accuracy of 86 per cent in a collection of 25.000 film reviews. The basic algorithm is designed to train O (n + V lg V) and O (n) for testing, where n is the number of words in the document (linear) and V is the size of the abbreviated dictionary. This is much faster than other machine learning algorithms, such as Maxent classification or support vector machines, which take a long time to get close to the optimal weight range. This accuracy is comparable to the accuracy of current algorithms used to identify moods in film reviews [13, 15].
So, you should start by defining a number of variables and grouping the data by class. As we can see from the performance, the best accuracy of ~86% was achieved on the test set.
recall presicion
Now let's look at other metrics such as accuracy, recall, and F1 score (formula 10) to take a critical look at the situation. This is why, instead of being reliable on its own, we generally refer to two other metrics: precision and recall and F1. Precision tests the percentage of the elements detected by the system (i.e. the system is labeled positive) that are actually positive (i.e. positive according to their gold labels). Precision is defined as (9) Recall measures the percentage of elements currently present in input data that have been correctly identified by the device. The callback is described as (8).
You can immediately see that Precision tells you how accurate/inaccurate your model is from those predicted positives, how many of them are actual positives. In our model 0.86.
In order to further assess the efficiency of the proposed pre-treatment stage, the outcomes of the previous and subsequent treatments are compared. However, if the results are worse than in the absence of a pre-processing period, which means that the classification model is not good enough, then changes are needed and the model is likely to be reconstructed. In addition, the naive classifier of Bayes will be checked with other classifiers (such as SVM, KNN) to show the superiority or refutation of the following: naive Bayes is better or at the same stage.
An F1 score is required when you need to find a balance between accuracy and recall. We have already shown that accuracy may be mainly attributed to a large number of real negatives, which, in most business situations, we do not concentrate on, though false-negative and positive generally have business costs (tangible and intangible), so the F1 score might be the best metric to be used if we need to find a balance between accuracy and response and the unequal distribution of classes (i.e. In this case, F1=0.86).
Bayes and SVM have been similar to each other, but the SVM approach has remained the leader in all measures of quantitative accuracy with qualitative indicators.
Method remained the leader in all indicators of objective accuracy with qualitative indicators.
The point of zero frequency is well known. You may use the anti-aliasing technique to solve this problem. One of the simplest smoothing techniques is the calculation of Laplace.
The presumption of independent predictors is another weakness of naive Bayes. In real life, it's almost difficult to get a set of predictors that are totally independent.
Thus, considering its unrealistic presumption of independence, the naive Bayesian classifier is surprisingly successful in practice, since its classification solution can often be right and its probability estimates accurate. As we have shown, even a very simple implementation of the naive Bayes algorithm can result in surprisingly good results for sentiment analysis. Notice that this model is basically a binary classifier, which means that it can be used for any dataset that has two categories. There are all sorts of applications for this, from spam detection to sentiment-based bitcoin trading.
The study shows that the SVM Classifier performs a better analysis of the accuracy of the above data sets compared to the commonly used KNN and Naive Bayes machine learning classifiers.
Both of the above analyzes help us foresee the arrival of goods on the market that could boost the income of the crushed organizations.
[12] D. Harel, Y. Koren, Clustering spatial data using random walks, Proc. of the 7th ACM SIGKDD Intern. conf. on knowledge discovery and data mining, San Francisco, California, 200, pp. 281–286. [13] N. Boyko, O. Pylypiv, Yu. Peleshchak, Yu. Kryvenchuk, J. Campos “Automated Document Analysis for Quick Personal Health Record Creation” The 2 nd International Workshop on Informatics & Data-Driven Medicine (IDDM 2019), Volume 1. Lviv, Ukraine, November 1113, 2019, pp. 208-221. [14] С. Zhang, Y. Murayama, “Testing local spatial autocorrelation using”, vol. 14, Intern. J. of
Geogr. Inform. Science, 2000, pp. 681–692. [15] N. Melnykova, V. Melnykov, E. Vasilevskis ”The personalized approach to the processing and analysis of patients' medical data”. CEUR Workshop Proceedings, 2018, Vol. 2255: Proceedings of the 1st International workshop on informatics & Data-driven medicine (IDDM 2018) Lviv, Ukraine, November 28–30, 2018., pp. 103-112. [16] V. Yakovyna, A. Peleshchyshyn, S. Albota ”Discussions of wikipedia talk pages: Manipulations detected by lingual-psychological analysis”, CEUR Workshop Proceedings, 2019, Vol. 2392, pp. 309-320.