CN103020249A - Classifier construction method and device as well as Chinese text sentiment classification method and system - Google Patents

Abstract

The invention provides a classifier construction method and device as well as a Chinese text sentiment classification method and system. The classification method comprises the following steps of: obtaining a sample to be labeled from a sample set to be labeled; looking up sentiment words in the sample to be labeled; obtaining the sentiment polarity of each sentiment word; converting the sentiment polarity of the sentiment words of which the sentiment polarity conforms to a sentiment polarity conversion rule in the sample to be labeled; counting the amount of the sentiment words of which the sentiment polarity is negative and positive in the sample to be labeled; according to the amount of the sentiment words of which the sentiment polarity is positive and the amount of the sentiment words of which the sentiment polarity is negative, determining the sentiment polarity of the sample to be labeled to obtain a labeled sample; according to the labeled sample, labeling other samples to be labeled in the sample set to be labeled to obtain a labeled sample set; constructing a maximum entropy classifier by the labeled sample set; and classifying a Chinese text to be classified by the maximum entropy classifier. According to the method, the device and the system provided by the invention, the Chinese text classification time is shortened, and the classification accuracy is improved.

Description

Classifier construction method and device, Chinese text sentiment classification method and system

technical field

The invention relates to the technical field of natural language processing and pattern recognition, in particular to a method and device for constructing a classifier, and a method and system for classifying Chinese text emotions.

Background technique

With the vigorous development of Web2.0, a large number of emotional comments on people, events, products, etc. have been produced on the Internet. Users can understand the opinions of public opinion on a certain event or product by browsing these comments. Due to the large amount of comment information, users will waste a lot of time and energy if they simply rely on manual collection and organization. Therefore, it is urgent to use computers to help users quickly obtain and organize these comment information. Text sentiment analysis technology came into being. born.

The so-called text sentiment analysis is the process of using computers to help users quickly acquire, organize and analyze comment information, and analyze, process, summarize and reason emotionally subjective texts. Among them, text sentiment classification is a basic content of text sentiment analysis. It can be divided into sentence level and article level according to different granularity. For sentence level and article level, text sentiment classification refers to dividing text into positive text and negative text. Text, for example, "I like this product very much", through text sentiment classification, this sentence will be classified as positive text, while "this book is in terrible" will be classified as negative text.

At present, the commonly used text sentiment classification method is based on the supervised method. This method uses the field-labeled data to train the classification of a specific field. , it takes a long time to build a classifier, and if you change a domain, you must relabel the corpus, that is, the domain dependence is large.

Contents of the invention

In view of this, the present invention provides a classifier construction method and device, a Chinese text sentiment classification method and system, to solve the problems that the existing classification method takes a long time to construct a classifier and has a large dependence on the application field . Its technical scheme is as follows:

A method for constructing a classifier, comprising:

Obtain a sample set to be labeled and obtain a sample to be labeled from the sample set to be labeled, wherein the sample set to be labeled includes at least two samples to be labeled;

Find the emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, wherein the emotional polarity includes positive and negative;

Change the emotional polarity of the emotional words that meet the emotional polarity conversion rule in the sample to be marked;

Counting the number of positive emotional words and emotional polarity in the sample to be marked is the number of negative emotional words;

Determine the emotional polarity of the sample to be marked according to the number of positive emotional words and negative emotional words according to the emotional polarity, and obtain the labeled sample;

Using a self-learning method to mark other samples to be marked in the sample set to be marked according to the marked sample, to obtain a set of marked samples;

A maximum entropy classifier is constructed by using labeled samples in the labeled sample set.

Preferably, converting the emotional polarity of the emotional words that meet the emotional polarity conversion rules in the sample to be marked includes:

If negative keywords appear in the sentence where the emotional word in the sample to be marked is located, then change the emotional polarity of the emotional word;

If there is a turning keyword in the next sentence or the next paragraph of the sentence where the emotional word in the sample to be labeled is, then change the emotional polarity of the emotional word;

And/or, if the sentence where the emotional word in the sample to be marked is located has a willing keyword, then change the emotional polarity of the emotional word.

Preferably, determining the emotional polarity of the sample to be marked according to the number of positive emotional words and negative emotional words according to the emotional polarity includes:

If the difference between the quantity of positive emotional words and the negative emotional words is greater than the set threshold, it is determined that the emotional polarity of the sample to be marked is positive;

If the difference between the number of emotional words whose emotional polarity is negative and the number of emotional words whose emotional polarity is positive is greater than the set threshold, it is determined that the emotional polarity of the sample to be labeled is negative.

Preferably, the emotional polarity of the sample to be marked is determined according to the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative, and the obtained labeled samples include:

Constructing a maximum entropy classifier using the labeled samples;

Using the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain a classification result;

The emotional polarity of each sample to be labeled is determined according to the classification result to obtain a labeled sample set.

A Chinese text sentiment classification method, comprising: the construction method of the above-mentioned classifier, also comprising:

The Chinese text to be classified is classified by using the constructed maximum entropy classifier.

A construction device for a classifier, comprising: an acquisition unit, a search unit, a polarity conversion unit, a statistical unit, a determination unit, a self-learning unit and a classifier construction unit;

The acquiring unit is configured to acquire a sample set to be labeled and acquire a sample to be labeled from the sample set to be labeled, wherein the sample set to be labeled includes at least two samples to be labeled;

The search unit is used to search for the emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, wherein the emotional polarity includes positive and negative;

The polarity conversion unit is used to convert the emotional polarity of the emotional words that meet the emotional polarity conversion rules in the sample to be marked;

The statistical unit is used to count the number of emotional words whose emotional polarity is positive and the emotional words whose emotional polarity is negative in the samples to be marked;

The determining unit is configured to determine the emotional polarity of the sample to be labeled according to the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative, to obtain a labeled sample;

The self-learning unit is configured to use a self-learning method to mark other samples to be marked in the set of samples to be marked according to the marked samples to obtain a set of marked samples;

The classifier construction unit is configured to use the labeled samples in the labeled sample set to construct a maximum entropy classifier.

Preferably, the polarity conversion unit includes: a first polarity conversion subunit, a second polarity conversion subunit and/or a third polarity conversion subunit;

The first polarity conversion subunit is used to convert the emotional polarity of the emotional word when negative keywords appear in the sentence where the emotional word in the sample to be marked is located;

The second polarity conversion subunit is used to change the emotional polarity of the emotional word when a turning keyword appears in the next sentence or the next paragraph of the sentence where the emotional word in the sample to be marked is located;

The third polarity conversion subunit is used to change the emotional polarity of the emotional word when the sentence where the emotional word in the sample to be marked is located has a willing keyword.

Preferably, the determination unit includes: a first determination subunit and a second determination subunit;

The first determining subunit is used to determine the emotional polarity of the sample to be marked when the difference between the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative is greater than a set threshold sex is positive;

The second determination subunit is used to determine the number of samples to be labeled when the difference between the number of emotional words whose emotional polarity is negative and the number of emotional words whose emotional polarity is positive is greater than the set threshold Emotional polarity is negative.

Preferably, the self-learning unit includes: a classifier construction subunit, a classification subunit and a third determination subunit;

The classifier construction subunit is used to construct a maximum entropy classifier using the labeled samples;

The classification subunit is configured to use the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain a classification result;

The third determination subunit is configured to determine the emotional polarity of each sample to be labeled according to the classification result.

A Chinese text emotion classification system, including the construction device of the above-mentioned classifier, also includes: a classification unit;

The classification unit is used to classify the Chinese text to be classified using the maximum entropy classifier constructed by the classifier construction device.

The construction method and device of the classifier provided by the present invention, the method and system of Chinese text emotion classification, apply the emotion polarity transformation rule to carry out the polarity transformation on the emotion polarity of the emotion device, and use the self-learning method to treat the labeling samples according to the labeling samples Collect other samples to be labeled for labeling, and use the maximum entropy classifier constructed according to the labeled samples of the labeled sample set as the classifier for Chinese text sentiment classification. The method and device for constructing a classifier provided by the present invention, the method and system for classifying emotion in Chinese texts, avoid the wasted cost of manual labeling of training samples, shorten the construction time of classifiers for sentiment classification in Chinese texts, and at the same time, improve The correct rate of Chinese text sentiment classification.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only It is an embodiment of the present invention, and those skilled in the art can also obtain other drawings according to the provided drawings without creative work.

FIG. 1 is a schematic flowchart of a method for constructing a classifier provided in an embodiment of the present invention;

FIG. 2 is a schematic structural diagram of a Chinese text sentiment classification system provided by an embodiment of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts belong to the protection scope of the present invention.

The present invention provides a method for constructing a classifier, and Fig. 1 is a flowchart of the method, and the method may include:

S101: Obtain a sample set to be labeled and obtain a sample to be labeled from the set of labeled samples, where the sample set to be labeled includes at least two samples to be labeled.

S102: Find the emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, where the emotional polarity of the emotional word includes positive and negative.

S103: Change the emotional polarity of the emotional words that meet the emotional polarity conversion rule in the samples to be labeled.

S104: Count the number of emotional words with positive emotional polarity and the number of emotional words with negative emotional polarity in the samples to be labeled.

S105: Determine the emotional polarity of the samples to be labeled according to the number of emotional words with positive emotional polarity and the number of emotional words with negative emotional polarity, and obtain labeled samples.

S106: Using a self-learning method to label other samples to be labeled in the sample set to be labeled according to the labeled sample to obtain a labeled sample set, the labeled sample set includes all labeled samples.

S107: Construct a maximum entropy classifier by using the labeled samples in the labeled sample set.

In another embodiment of the present invention, step S102 may include: comparing the preset correspondence table of emotional words and emotional polarities, searching for emotional words in the samples to be labeled, and obtaining the emotional polarity corresponding to the emotional words according to the corresponding table. sex. Table 1 provides a table of correspondence between emotional words and emotional polarities. It should be noted that this embodiment is not limited to the emotional words listed in Table 1, and there may be other emotional words.

Table 1

emotional polarity emotional words front like, happy, satisfied, good, good the negative hate, bored, sad, bad

In yet another embodiment of the present invention, the emotion polarity transition rules may include: negation rules, turning rules and/or aspirational rules. Correspondingly, step S103 may include: if negative keywords appear in the sentence where the emotional word in the sample to be marked is located, then change the emotional polarity of the emotional word; if the next sentence of the sentence where the emotional word in the sample to be marked is located Or if there is a turning keyword in the next paragraph, then change the emotional polarity of the emotional word; and/or, if there is a willing keyword in the sentence where the emotional word in the sample to be marked is, then change the emotional polarity of the emotional word . Table 2 lists common negative keywords, turning keywords and willing keywords. Of course, this embodiment is not limited to these keywords, and may include other keywords representing negation, turning and willing.

Table 2

The following is an illustration of three specific examples of changing the emotional polarity of emotional words based on the negation rule, the turning rule and the willingness rule:

Example 1: I don't like this product.

In the sentence of example 1, if the emotional word is "like", and the negative keyword "no" appears in the sentence, then change the emotional polarity of the emotional word "like", that is, change the emotional polarity of "like" from positive to positive turned negative.

Example 2: I like the idea of this product, but the quality is not acceptable to me.

In the sentence of example 2, if the emotional word is "like", and the transition keyword "but" appears in the next sentence of the sentence, then change the emotional polarity of the emotional word "like", that is, the emotion of "like" The polarity changes from positive to negative.

Example 3: It would be fine if the color was red.

In the sentence of example 3, if the emotional word is "good", and in the sentence where it is located, the keyword "if" appears before the emotional word "good", then change the emotional polarity of the emotional word "good" , which is to change the emotional polarity of "good" from positive to negative.

In yet another embodiment of the present invention, step S106 may include: if the difference between the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative is greater than a set threshold, then determine the samples to be labeled The emotional polarity of is positive; if the difference between the number of emotional words whose emotional polarity is negative and the number of emotional words whose emotional polarity is positive is greater than the set threshold, then it is determined that the emotional polarity of the sample to be labeled is negative. Assume that the number of emotional words with positive emotional polarity is N ₊ , and the number of emotional words with negative emotional polarity is N _- , set the threshold to N _max , if N ₊ -N _- > N _max , determine to be labeled The emotional polarity of the sample is positive, and if N _- -N ₊ >N _max , it is determined that the emotional polarity of the sample to be labeled is negative.

In yet another embodiment of the present invention, step S105 may include: using the labeled samples to construct a maximum entropy classifier; using the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain classification results, and determine each The emotional polarity of each sample to be labeled, and finally get two standard sample sets: a positive label sample set and a negative label sample set.

Among them, the maximum entropy classifier, as one of the machine learning classification methods, is based on the maximum entropy information theory, and its basic idea is to build a model for all known factors and exclude all unknown factors. That is, to find a probability distribution that satisfies all known facts but makes the unknown factors most random. Compared with the naive Bayesian method, the biggest feature of this method is that it does not need to satisfy the conditional independence between features. Therefore, this method is suitable for fusing various features without considering the influence between them.

Under the maximum entropy model, the formula for predicting the conditional probability P(c|D) is as follows:

$\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; |</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; 1</span>}}{\mathrm{<span\; class="notranslate">\; Z</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; )</span>}\mathrm{<span\; class="notranslate">\; exp</span>}<span\; class="notranslate">\; (</span>\underset{\mathrm{<span\; class="notranslate">\; k</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; c</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span>$

where Z(D) is the normalization factor. P _k,c is the characteristic function, defined as:

${\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; 1</span>}& {\mathrm{<span\; class="notranslate">\; no</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; 0</span>}\mathrm{<span\; class="notranslate">\; and</span>}{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; c</span>}\\ \mathrm{<span\; class="notranslate">\; 0</span>}& \mathrm{<span\; class="notranslate">\; oterwise</span>}\end{array}\right.$

The present invention also provides a method for sentiment classification of Chinese texts. In addition to the above steps S101-S107, the method also includes: classifying the Chinese texts to be classified by using the constructed maximum entropy classifier.

In order to compare the Chinese text sentiment classification method provided in this example with the existing Chinese text sentiment classification method, this example uses comment corpus in some fields as unmarked samples to be classified, and tests the two classification methods respectively . The corpus used in the test is data from two domains, namely reviews about bags and hotels. The evaluation standard selected in the experiment is Accuracy, which is a comprehensive evaluation standard for evaluating general classification problems. For each field, the standard rate is calculated as Accuracy=(TP+NP)/A, where TP refers to the correct classification of positive text The total number of samples, NP refers to the total number of samples with correct negative text classification, and A refers to the total number of samples with correct classification selected.

It should be noted that the correctness of the sentiment polarity of the Chinese text to be classified is judged. The specific content of the judgment is that, in the positive text, if the number of positive emotional words is more than the number of negative emotional words, the classification of the positive text is correct; In the text, if the number of positive emotional words is less than or equal to the number of negative emotional words, the positive text classification is incorrect; in the negative text, if the number of negative emotional words is more than the number of positive emotional words, the negative text classification is correct; in the negative text In the text, if the number of negative sentiment words is less or equal to the number of positive sentiment words, the classification of negative text is incorrect.

Table 3 compares the result of classifying Chinese texts for adopting the classification method provided by the present invention and adopting the classification method of the prior art:

table 3

In this experiment, different numbers of labeled samples were used for experimental verification, and N _max =3.

The traditional classification method will calculate the number of emotional words in each sample as the basis for determining the emotional category of the sample. The method that the embodiment of the present invention provides, first adopts emotion polarity transformation rule to emotion word and has done polarity transformation judgment to emotion word, and rule comprises: negation rule, transition rule, can willing rule, has avoided emotion polarity transformation to emotion word The impact of the judgment, and the maximum entropy classifier constructed after automatic labeling of non-labeled samples is used for Chinese text sentiment classification.

As can be seen from the data in Table 3, the accuracy rate of the Chinese emotion classification method provided by the present embodiment is much higher than that of the traditional text emotion classification method, and the rate of improvement can exceed 3 percentage points at most. Again, It is proved that the classification method provided by this embodiment has a high accuracy rate, reduces the cost of manual labeling, and avoids the negative impact of emotional words with emotional polarity changes on the text classification results, which is conducive to improving the text classification effect.

Corresponding to the construction method of the above-mentioned classifier, the embodiment of the present invention also provides a construction device of the classifier. FIG. 103 , a statistical unit 104 , a determination unit 105 , a self-learning unit 106 and a classifier construction unit 107 . in:

The acquiring unit 101 is configured to acquire a sample set to be labeled and acquire a sample to be labeled from the sample set to be labeled, wherein the sample set to be labeled includes at least two samples to be labeled. The searching unit 102 is configured to search for emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, wherein the emotional polarity includes positive and negative. The polarity conversion unit 103 is configured to convert the emotional polarity of the emotional words in the sample to be labeled that meet the emotional polarity conversion rule. The statistical unit 104 is configured to count the number of emotional words with positive emotional polarity and the number of emotional words with negative emotional polarity in the samples to be labeled. The determining unit 105 is configured to determine the emotional polarity of the sample to be labeled according to the number of emotional words with positive emotional polarity and the number of emotional words with negative emotional polarity, so as to obtain labeled samples. The self-learning unit 106 is configured to use a self-learning method based on the labeled samples to label other samples to be labeled in the sample set to be labeled to obtain a set of labeled samples. A classifier construction unit 107, configured to construct a maximum entropy classifier by using the labeled samples in the labeled sample set.

In another embodiment of the present invention, the polarity conversion unit 103 may include: a first polarity conversion subunit, a second polarity conversion subunit and/or a third polarity conversion subunit. in:

The first polarity conversion subunit is used to change the emotional polarity of the emotional word when negative keywords appear in the sentence where the emotional word in the sample to be labeled is located. The second polarity conversion subunit is used to change the emotional polarity of the emotional word when a turning keyword appears in the next sentence or paragraph of the sentence where the emotional word in the sample to be labeled is located. The third polarity conversion subunit is used to change the emotional polarity of the emotional word when the sentence where the emotional word in the sample to be marked is located has a willing keyword.

In another embodiment of the present invention, the determining unit 105 may include: a first determining subunit and a second determining subunit. in:

The first determining subunit is used to determine that the emotional polarity of the sample to be labeled is positive when the difference between the number of emotional words with positive emotional polarity and the number of emotional words with negative emotional polarity is greater than a set threshold. The second determination subunit is used to determine that the emotional polarity of the sample to be labeled is negative when the difference between the number of emotional words with negative emotional polarity and the number of positive emotional words with emotional polarity is greater than a set threshold.

In yet another embodiment of the present invention, the self-learning unit 106 may include: a classifier construction subunit, a classification subunit, and a third determination subunit. in:

The classifier construction subunit is used to construct a maximum entropy classifier using labeled samples. The classification subunit is used to use the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain a classification result. The third determining subunit is configured to determine the emotional polarity of each sample to be labeled according to the classification result.

Corresponding to the above-mentioned Chinese text sentiment classification method, an embodiment of the present invention also provides a Chinese text sentiment classification system, the system further includes: a classification unit in addition to the above-mentioned classifier construction device. Wherein, the classification unit is used to classify the Chinese text to be classified using the maximum entropy classifier constructed by the classifier construction device.

The method and device for constructing a classifier provided by the embodiments of the present invention, the method and system for classifying Chinese text emotions, apply the emotion polarity transformation rule to carry out polarity transformation on the emotion polarity of the emotion device, and use the method of self-learning according to the labeled samples to treat The other samples to be labeled in the labeled sample set are labeled, and the maximum entropy classifier constructed based on the labeled samples in the labeled sample set is used as the classifier for Chinese text sentiment classification. The method and device for constructing a classifier provided by the present invention, the method and system for classifying emotion in Chinese texts, avoid the wasted cost of manual labeling of training samples, shorten the construction time of classifiers for sentiment classification in Chinese texts, and at the same time, improve The correct rate of Chinese text sentiment classification.

For the convenience of description, when describing the above devices, functions are divided into various units and described separately. Of course, when implementing the present invention, the functions of each unit can be implemented in one or more pieces of software and/or hardware.

It can be seen from the above description of the implementation manners that those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in storage media, such as ROM/RAM, disk , CD, etc., including several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods described in various embodiments or some parts of the embodiments of the present invention.

Each embodiment in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for related parts, please refer to part of the description of the method embodiment. The system embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without creative effort.

The invention is applicable to numerous general purpose and special purpose computing system environments or configurations. Examples: personal computers, server computers, handheld or portable devices, tablet-type devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, including A distributed computing environment for any of the above systems or devices, etc.

The invention may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including storage devices.

It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is a relationship between these entities or operations. There is no such actual relationship or order between them.

The foregoing is only a specific embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art, some improvements and modifications can also be made without departing from the principle of the present invention. It should be regarded as the protection scope of the present invention.

Claims (10)

1. A method for building a classifier, comprising: Obtain a sample set to be labeled and obtain a sample to be labeled from the sample set to be labeled, wherein the sample set to be labeled includes at least two samples to be labeled; Find the emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, wherein the emotional polarity includes positive and negative; Change the emotional polarity of the emotional words that meet the emotional polarity conversion rule in the sample to be marked; Counting the number of positive emotional words and emotional polarity in the sample to be marked is the number of negative emotional words; Determine the emotional polarity of the sample to be marked according to the number of positive emotional words and negative emotional words according to the emotional polarity, and obtain the labeled sample; Using a self-learning method to mark other samples to be marked in the sample set to be marked according to the marked sample, to obtain a set of marked samples; A maximum entropy classifier is constructed by using labeled samples in the labeled sample set. 2. method according to claim 1, is characterized in that, transforming the emotional polarity that meets the emotional polarity conversion rule of emotional polarity in the sample to be labeled comprises: If negative keywords appear in the sentence where the emotional word in the sample to be marked is located, then change the emotional polarity of the emotional word; If there is a turning keyword in the next sentence or the next paragraph of the sentence where the emotional word in the sample to be labeled is, then change the emotional polarity of the emotional word; And/or, if the sentence where the emotional word in the sample to be marked is located has a willing keyword, then change the emotional polarity of the emotional word. 3. method according to claim 1, is characterized in that, described according to described emotional polarity is positive emotional word quantity and emotional polarity is negative emotional word quantity and determines the emotional extreme of described sample to be labeled Sex includes: If the difference between the quantity of positive emotional words and the negative emotional words is greater than the set threshold, it is determined that the emotional polarity of the sample to be marked is positive; If the difference between the number of emotional words whose emotional polarity is negative and the number of emotional words whose emotional polarity is positive is greater than the set threshold, it is determined that the emotional polarity of the sample to be labeled is negative. 4. method according to claim 1, is characterized in that, utilizes the method for self-learning according to described labeled sample to label other samples to be labeled in described sample set to be labeled, obtain labeled sample set comprising: Constructing a maximum entropy classifier using the labeled samples; Using the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain a classification result; The emotional polarity of each sample to be labeled is determined according to the classification result to obtain a labeled sample set. 5. A Chinese text emotion classification method, is characterized in that, comprises: the construction method of the classifier as described in any one in claim 1-4, also comprises: The Chinese text to be classified is classified by using the constructed maximum entropy classifier. 6. A construction device for a classifier, characterized in that it comprises: an acquisition unit, a search unit, a polarity conversion unit, a statistical unit, a determination unit, a self-learning unit and a classifier construction unit; The acquiring unit is configured to acquire a sample set to be labeled and acquire a sample to be labeled from the sample set to be labeled, wherein the sample set to be labeled includes at least two samples to be labeled; The search unit is used to search for the emotional words in the sample to be labeled, and obtain the emotional polarity of each emotional word, wherein the emotional polarity includes positive and negative; The polarity conversion unit is used to convert the emotional polarity of the emotional words that meet the emotional polarity conversion rules in the sample to be marked; The statistical unit is used to count the number of emotional words whose emotional polarity is positive and the emotional words whose emotional polarity is negative in the samples to be marked; The determining unit is configured to determine the emotional polarity of the sample to be labeled according to the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative, to obtain a labeled sample; The self-learning unit is configured to use a self-learning method to mark other samples to be marked in the set of samples to be marked according to the marked samples to obtain a set of marked samples; The classifier construction unit is configured to use the labeled samples in the labeled sample set to construct a maximum entropy classifier. 7. The device according to claim 6, wherein the polarity conversion unit comprises: a first polarity conversion subunit, a second polarity conversion subunit and/or a third polarity conversion subunit; The first polarity conversion subunit is used to convert the emotional polarity of the emotional word when negative keywords appear in the sentence where the emotional word in the sample to be marked is located; The second polarity conversion subunit is used to change the emotional polarity of the emotional word when a turning keyword appears in the next sentence or the next paragraph of the sentence where the emotional word in the sample to be marked is located; The third polarity conversion subunit is used to change the emotional polarity of the emotional word when the sentence where the emotional word in the sample to be marked is located has a willing keyword. 8. The device according to claim 6, wherein the determining unit comprises: a first determining subunit and a second determining subunit; The first determining subunit is used to determine the emotional polarity of the sample to be marked when the difference between the number of emotional words whose emotional polarity is positive and the number of emotional words whose emotional polarity is negative is greater than a set threshold sex is positive; The second determination subunit is used to determine the number of samples to be labeled when the difference between the number of emotional words whose emotional polarity is negative and the number of emotional words whose emotional polarity is positive is greater than the set threshold Emotional polarity is negative. 9. The device according to claim 6, wherein the self-learning unit comprises: a classifier construction subunit, a classification subunit and a third determination subunit; The classifier construction subunit is used to construct a maximum entropy classifier using the labeled samples; The classification subunit is configured to use the maximum entropy classifier to classify other samples to be labeled in the sample set to be labeled to obtain a classification result; The third determination subunit is configured to determine the emotional polarity of each sample to be labeled according to the classification result. 10. A Chinese text emotion classification system is characterized in that, comprising the construction device of the classifier as described in any one of claims 6-9, also comprising: classification unit; The classification unit is used to classify the Chinese text to be classified by using the maximum entropy classifier constructed by the classifier construction device.

Images