CN105138510A - Microblog-based neologism emotional tendency judgment method - Google Patents

Abstract

The invention relates to a microblog-based neologism emotional tendency judgment method, belonging to the field of natural language processing. The microblog-based neologism emotional tendency judgment method disclosed by the invention comprises the following steps: dividing words of microblog corpuses through a Chinese word division tool, blocking the corpuses, the words in which are divided, by taking stop words in a word division result as a division point, pairwise combining adjacent word strings in each block, calculating the combined word string frequency, and taking the word strings, the frequencies of which are higher than a threshold value, as neologism candidate strings; filtering the neologism candidate strings according to a word formation rule of Chinese linguistics and an adjacent change number rule so as to obtain neologisms; calculating the similarity between co-occurrence words and hownet emotional words by utilizing an emotional dictionary of a hownet; calculating the relevancy between the neologisms and the co-occurrence words; constructing an image model; and obtaining the emotional polarity distribution of the neologisms by utilizing a label propagation algorithm, and obtaining the emotional tendency of the neologisms by constructing a linear classifier. By means of judgement of the emotional tendency of the neologisms, a blogger can express views better; and furthermore, the emotional tendency of the blogger can be accurately known by users.

Description

A kind of neologisms Sentiment orientation decision method based on microblogging

Technical field

The present invention relates to a kind of neologisms Sentiment orientation decision method based on microblogging, belong to natural language processing field.

Background technology

A large amount of emotion neologisms emerge in large numbers in microblogging, and appearing in the daily interchange of people of these neologisms plays an important role, can the abundanter comprehensive expression viewpoint of people and emotion, the simultaneously refraction of Ye Shi social trend and media event.In natural language processing process, emotion new word identification is a difficulties always, and it has very important application in Chinese word segmentation, information retrieval, question answering system etc.

Current vocabulary feeling polarities recognition methods, first choose there is intense emotion tendency word as benchmark word, then determine the feeling polarities of target word by the strength of association calculated between benchmark word.TurneyPD. wait people to use PMI-IR method, utilization point mutual information represents the strength of association between target word and benchmark word, obtains the feeling polarities of target word; The people such as Wang Suge use PMI method, calculate word and synonym thereof respectively and pass judgement on the strength of association of benchmark word set, then judge the feeling polarities of word according to the difference of strength of association.The people such as Li Dun think that the vocabulary of co-occurrence has identical feeling polarities.Utilize " good ", " green bristlegrass " polarity justice in HowNet former, the polarity similarity that calculating word and benchmark word are anticipated between item, thus calculate word polarity number.Yao Tian Fang, Wan Changxuan etc. introduce link word (but and etc.) when calculating word association intensity, and utilize word part of speech within a context and syntactic structure information, calculate dynamic (modification) polarity of word.Improve the accuracy that feeling polarities calculates.

For the feeling polarities identification of neologisms, neologisms lack the priori of part of speech, semantic aspect on the one hand, cannot directly utilize the feeling polarities to neologisms such as external resource of knowing net to judge; On the other hand because the quantity of neologisms and benchmark word is all relatively limited, only calculate the degree of correlation with benchmark word, there will be serious Sparse Problem.The emotion neologisms polarity identification method based on microblogging language material that the present invention proposes is when calculating neologisms feeling polarities, not only consider the benchmark word be associated with neologisms, and consider the non-referenced word and other neologisms with Sentiment orientation, on the impact of neologisms feeling polarities identification.

Summary of the invention

The invention provides a kind of neologisms Sentiment orientation decision method based on microblogging, the problem that the emotion neologisms in existing situation in microblogging language material cannot identify automatically can be solved.

Technical scheme of the present invention is: a kind of neologisms Sentiment orientation decision method based on microblogging, by Chinese word segmentation instrument, participle is carried out to microblogging language material, and with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate; According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms; Recycling knows the sentiment dictionary of net, calculates co-occurrence word and the Words similarity knowing net emotion word; Calculate the degree of correlation of neologisms and co-occurrence word; With the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation of neologisms and co-occurrence word, design of graphics model; Utilize label propagation algorithm to obtain the feeling polarities distribution of neologisms, obtain the emotion tendency of neologisms finally by structure linear classifier.

The concrete steps of described method are as follows:

Step1, by Chinese word segmentation instrument, participle is carried out to microblogging language material;

Step2, with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, and by word string combination of two adjacent in each piece, the word string frequency after combination is added up, the word string of frequency higher than threshold value T is gone here and there as neologisms candidate;

Step3, rule that is regular according to the one-tenth word of Chinese and adjacent change number are carried out filtration to neologisms candidate string and are obtained neologisms;

Step4, utilize and know the sentiment dictionary of net, calculate co-occurrence word and the Words similarity knowing net emotion word:

Step4.1, find out the non-stop words with new Term co-occurrence in microblogging language material, as co-occurrence word;

Net sentiment dictionary is known in Step4.2, utilization, calculates co-occurrence word and the Words similarity knowing net emotion word, is expressed as follows:

In formula, Sim (s _i, p _j) represent co-occurrence word s _iwith know net emotion word p _jbetween Words similarity, i and j represents the subscript of arbitrary two words, m and n is respectively co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry number, represent co-occurrence word s _im the senses of a dictionary entry, represent and know net emotion word p _jn-th senses of a dictionary entry, P represents the set knowing net emotion word;

$sim(u_m^{s_i},u_n^{p_j})=\frac{1}{n_1{n}_2}\underset{l=1}{\overset{n_1}{\Σ}}\underset{f=1}{\overset{n_2}{\Σ}}\frac{1}{2^{n_1-l+1}}sim(r^{s_i},r^{p_j})$

In formula, represent co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry similarity, n ₁and n ₂be respectively the senses of a dictionary entry with in attribute number, for the weighted value of the attribute of diverse location in senses of a dictionary entry definition, l is 1 to n ₁a variable, f is 1 to n ₂a variable, with for the senses of a dictionary entry with justice unit;

In formula, represent co-occurrence word s _iwith know net emotion word p _jjustice unit similarity, d is with path distance in hierarchical system, α is an adjustable parameter;

The degree of correlation of the co-occurrence word of Step5, calculating neologisms and neologisms:

$w_{ij}=\underset{r=0}{\overset{R}{\Σ}}C(v_i,v_j)\×N(v_i,r,v_j)$

In formula, i and j represents the subscript of arbitrary two words, and R is the size of self-defining window, and r is the positive number being less than or equal to R, represents the distance of two words in R window, w _ijrepresent neologisms v _iwith neologisms v _ico-occurrence word v _jthe degree of correlation, N (v _i, r, v _j) be: neologisms v _iwith neologisms v _ico-occurrence word .v _j. the co-occurrence number of times (r < R) when distance is r in R window in relevant documentation set, C (v _i, v _j)=R-r+1 is the co-occurrence intensity between two words;

Step6, with the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation in Step5, design of graphics model;

Step7, utilize Word similarity co-occurrence word s _ipolarity distribution on y:

$s_i\left(y\right)=\{\begin{array}{c}\frac{count\left(Sim\right(s_i,p_j^{+})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=+1\\ 0\\ \frac{count\left(Sim\right(s_i,p_j^{-})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=-1\end{array}y=0$

In formula, RK is threshold value, s _iy () represents the polarity distribution of co-occurrence word, i and j represents the subscript of arbitrary two words, and Sim represents Words similarity, for knowing just tendentious word in net emotion word, for knowing in net emotion word negative tendentious word, the quantity of what count represented is word;

Step8, label propagation algorithm determination neologisms Sentiment orientation;

Step8.1, obtain neologisms polarity distribution, its objective function is as follows:

In the polarity identification of neologisms, think that the degree of correlation between word is higher, the polarity distribution between them is more similar.Based on above thought, set up following objective function.Make objective function C minimum, and then obtain neologisms node v _ipolarity distribution.

$C=\underset{\underset{v_j\&Element;K\left(V_t\right)}{v_i\&Element;V_t}}{\&Sigma;}{w}_{ij}\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2+\mathrm{\&lambda;}\underset{v_i\&Element;V_t}{\&Sigma;}||q_i\left(y\right)-\mathrm{\&gamma;}|{|}^2$

Wherein:

$\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-s_i(y\left)\right)}^2$

$\left|\right|q_i\left(y\right)-\mathrm{\&gamma;}|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-\mathrm{\&gamma;}(y\left)\right)}^2$

In formula, i and j represents the subscript of arbitrary two words, q _iy () represents neologisms node v _ipolarity distribution, s _iy () represents co-occurrence word v _jpolarity distribution, γ and λ is custom parameter, V _trepresent co-occurrence word set, K (V _t) represent co-occurrence word V _tk nearest neighbor set of words;

Step8.2, by the polarity of the neologisms obtained distribution be designated as Q _n, build linear classifier, obtain the Sentiment orientation of neologisms: when the probability that the Sentiment orientation of neologisms the is commendation probability deducted as derogatory sense is greater than threshold value RT, this neologisms Sentiment orientation is 1, is namely commendatory term; When the probability that the Sentiment orientation of neologisms the is commendation absolute value deducted as the probability of derogatory sense is less than threshold value RT, this neologisms Sentiment orientation is 0, is namely neutral words; When the probability that the Sentiment orientation of neologisms the is derogatory sense probability deducted as commendation is greater than threshold value RT, this neologisms Sentiment orientation is-1, is namely derogatory term;

$Q_n=\left\{\begin{array}{cc}1& Q_n(y=1)-Q_n(y=-1)>RT\\ 0& |Q_n(y=1)-Q_n(y=-1)|\&le;RT\\ -1& Q_n(y=-1)-Q_n(y=1)>RT\end{array}\right.$

In formula, Q _n(y=1) Q is represented _nfor the probability of commendation, Q _n(y=-1) Q is represented _nfor the probability of derogatory sense, RT is threshold value.

Principle of work of the present invention is:

Step 01) first obtain microblogging neologisms:

By Chinese word segmentation instrument, participle is carried out to microblogging language material.With the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, the word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate.According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms;

Step 02) extract the emotion word knowing net, find out the word with new Term co-occurrence in microblogging language material.By these words and the similarity knowing their senses of a dictionary entry of Similarity Measure that the justice of net emotion word is former, and then obtain the similarity of word according to the similarity of the senses of a dictionary entry.Calculate the probability distribution of co-occurrence word polarity, select have intense emotion tendency co-occurrence word and neologisms as node, according to word and co-occurrence word and and other neologisms between co-occurrence Strength co-mputation node between the degree of correlation, as the weight on limit between node, design of graphics model;

Step 03) method that adopts label to propagate, known feeling polarities is distributed, is delivered to the neologisms strong with its degree of correlation, the distribution of the neologisms polarity of the unknown is calculated.Passing judgement on polarity distribution spatially according to neologisms, building linear classifier, the feeling polarities of neologisms is identified.

The invention has the beneficial effects as follows: a large amount of neologisms occur along with the fast development of microblogging, bloger can not only be made well to express the viewpoint of oneself to the judgement of the emotion tendency of these neologisms, and user can be allowed to hold bloger's Sentiment orientation accurately.Meanwhile, judge the Sentiment orientation of these neologisms in time, have very important meaning to the word segmentation processing in Chinese information processing, the analysis of public opinion etc.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention.

Embodiment

Embodiment 1: as shown in Figure 1, a kind of neologisms Sentiment orientation decision method based on microblogging, by Chinese word segmentation instrument, participle is carried out to microblogging language material, and with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate; According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms; Recycling knows the sentiment dictionary of net, calculates co-occurrence word and the Words similarity knowing net emotion word; Calculate the degree of correlation of neologisms and co-occurrence word; With the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation of neologisms and co-occurrence word, design of graphics model; Utilize label propagation algorithm to obtain the feeling polarities distribution of neologisms, obtain the emotion tendency of neologisms finally by structure linear classifier.

The concrete steps of described method are as follows:

Step1, by Chinese word segmentation instrument, participle is carried out to microblogging language material;

Step2, with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, and by word string combination of two adjacent in each piece, the word string frequency after combination is added up, the word string of frequency higher than threshold value T is gone here and there as neologisms candidate;

Step3, rule that is regular according to the one-tenth word of Chinese and adjacent change number are carried out filtration to neologisms candidate string and are obtained neologisms;

Step4, utilize and know the sentiment dictionary of net, calculate co-occurrence word and the Words similarity knowing net emotion word:

Step4.1, find out the non-stop words with new Term co-occurrence in microblogging language material, as co-occurrence word;

Net sentiment dictionary is known in Step4.2, utilization, calculates co-occurrence word and the Words similarity knowing net emotion word, is expressed as follows:

In formula, Sim (s _i, p _j) represent co-occurrence word s _iwith know net emotion word p _jbetween Words similarity, i and j represents the subscript of arbitrary two words, m and n is respectively co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry number, represent co-occurrence word s _im the senses of a dictionary entry, represent and know net emotion word p _jn-th senses of a dictionary entry, P represents the set knowing net emotion word;

$sim(u_m^{s_i},u_n^{p_j})=\frac{1}{n_1{n}_2}\underset{l=1}{\overset{n_1}{\&Sigma;}}\underset{f=1}{\overset{n_2}{\&Sigma;}}\frac{1}{2^{n_1-l+1}}sim(r^{s_i},r^{p_j})$

In formula, represent co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry similarity, n ₁and n ₂be respectively the senses of a dictionary entry with in attribute number, for the weighted value of the attribute of diverse location in senses of a dictionary entry definition, l is 1 to n ₁a variable, f is 1 to n ₂a variable, with for the senses of a dictionary entry with justice unit;

In formula, represent co-occurrence word s _iwith know net emotion word p _jjustice unit similarity, d is with path distance in hierarchical system, α is an adjustable parameter;

The degree of correlation of the co-occurrence word of Step5, calculating neologisms and neologisms:

$w_{ij}=\underset{r=0}{\overset{R}{\&Sigma;}}C(v_i,v_j)\&times;N(v_i,r,v_j)$

In formula, i and j represents the subscript of arbitrary two words, and R is the size of self-defining window, and r is the positive number being less than or equal to R, represents the distance of two words in R window, w _ijrepresent neologisms v _iwith neologisms v _ico-occurrence word v _jthe degree of correlation, N (v _i, r, v _j) be: neologisms v _iwith neologisms v _ico-occurrence word v _jco-occurrence number of times (r < R) when distance is r in R window in relevant documentation set, C (v _i, v _j)=R-r+1 is the co-occurrence intensity between two words;

Step6, with the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation in Step5, design of graphics model;

Step7, utilize Word similarity co-occurrence word si on y polarity distribution:

$s_i\left(y\right)=\{\begin{array}{c}\frac{count\left(Sim\right(s_i,p_j^{+})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=+1\\ 0\\ \frac{count\left(Sim\right(s_i,p_j^{-})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=-1\end{array}y=0$

In formula, RK is threshold value, s _iy () represents the polarity distribution of co-occurrence word, i and j represents the subscript of arbitrary two words, and Sim represents Words similarity, for knowing just tendentious word in net emotion word, for knowing in net emotion word negative tendentious word, the quantity of what count represented is word;

Step8, label propagation algorithm determination neologisms Sentiment orientation;

Step8.1, obtain neologisms polarity distribution, its objective function is as follows:

$C=\underset{\underset{v_j\&Element;K\left(V_t\right)}{v_i\&Element;V_t}}{\&Sigma;}{w}_{ij}\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2+\mathrm{\&lambda;}\underset{v_i\&Element;V_t}{\&Sigma;}||q_i\left(y\right)-\mathrm{\&lambda;}|{|}^2$

Wherein:

$\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-s_i(y\left)\right)}^2$

$\left|\right|q_i\left(y\right)-\mathrm{\&gamma;}|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-\mathrm{\&gamma;}(y\left)\right)}^2$

In formula, i and j represents the subscript of arbitrary two words, q _iy () represents neologisms node v _ipolarity distribution, s _iy () represents co-occurrence word v _jpolarity distribution, γ and λ is custom parameter, V _trepresent co-occurrence word set, K (V _t) represent co-occurrence word V _tk nearest neighbor set of words;

Step8.2, by the polarity of the neologisms obtained distribution be designated as Q _n, build linear classifier, obtain the Sentiment orientation of neologisms: when the probability that the Sentiment orientation of neologisms the is commendation probability deducted as derogatory sense is greater than threshold value RT, this neologisms Sentiment orientation is 1, is namely commendatory term; When the probability that the Sentiment orientation of neologisms the is commendation absolute value deducted as the probability of derogatory sense is less than threshold value RT, this neologisms Sentiment orientation is 0, is namely neutral words; When the probability that the Sentiment orientation of neologisms the is derogatory sense probability deducted as commendation is greater than threshold value RT, this neologisms Sentiment orientation is-1, is namely derogatory term;

$Q_n=\left\{\begin{array}{cc}1& Q_n(y=1)-Q_n(y=-1)>RT\\ 0& |Q_n(y=1)-Q_n(y=-1)|\&le;RT\\ -1& Q_n(y=-1)-Q_n(y=1)>RT\end{array}\right.$

In formula, Q _n(y=1) Q is represented _nfor the probability of commendation, Q _n(y=-1) Q is represented _nfor the probability of derogatory sense, RT is threshold value.

Embodiment 2: as shown in Figure 1, a kind of neologisms Sentiment orientation decision method based on microblogging, by Chinese word segmentation instrument, participle is carried out to microblogging language material, and with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate; According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms; Recycling knows the sentiment dictionary of net, calculates co-occurrence word and the Words similarity knowing net emotion word; Calculate the degree of correlation of neologisms and co-occurrence word; With the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation of neologisms and co-occurrence word, design of graphics model; Utilize label propagation algorithm to obtain the feeling polarities distribution of neologisms, obtain the emotion tendency of neologisms finally by structure linear classifier.

Embodiment 3: as shown in Figure 1, a kind of neologisms Sentiment orientation decision method based on microblogging, by Chinese word segmentation instrument, participle is carried out to microblogging language material, and with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate; According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms; Recycling knows the sentiment dictionary of net, calculates co-occurrence word and the Words similarity knowing net emotion word; Calculate the degree of correlation of neologisms and co-occurrence word; With the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation of neologisms and co-occurrence word, design of graphics model; Utilize label propagation algorithm to obtain the feeling polarities distribution of neologisms, obtain the emotion tendency of neologisms finally by structure linear classifier.

The concrete steps of described method are as follows:

Step1, utilize Chinese Academy of Sciences's participle instrument, participle is carried out to microblogging language material, is input as 1,000 ten thousand microblogging language materials herein, export as microblogging language material after participle;

Step2, obtain neologisms candidate string, be input as microblogging language material after participle herein, export as neologisms candidate string:

With the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two (combined situation is as shown in table 1) adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value T=8 is gone here and there as neologisms candidate;

Word combination between table 1 stop words

Table 1 is depicted as the language material after participle, by the process of the word string combination of two between adjacent stop words.

Step3, acquisition neologisms, be input as neologisms candidate string herein, export as neologisms:

First filter neologisms candidate string according to the one-tenth word rule of Chinese, secondly utilize the rule of adjacent change number to carry out filtration to neologisms candidate string and obtain neologisms, the generative process of neologisms and number change are as table 2:

Table 2 neologisms generative process and number change

Table 2 is depicted as the change of the quantity of the word of every one-phase in new word discovery process.

Step4, utilize and know the sentiment dictionary of net, calculate co-occurrence word and the Words similarity knowing net emotion word:

Step4.1, find out the non-stop words with new Term co-occurrence in microblogging language material, as co-occurrence word;

Net sentiment dictionary is known in Step4.2, utilization, calculates co-occurrence word and the Words similarity knowing net emotion word, is expressed as follows:

In formula, Sim (s _i, p _j) represent co-occurrence word s _iwith know net emotion word p _jbetween Words similarity, i and j represents the subscript of arbitrary two words, m and n is respectively co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry number, represent co-occurrence word s _im the senses of a dictionary entry, represent and know net emotion word p _jn-th senses of a dictionary entry, P represents the set knowing net emotion word;

$\mathrm{si}m(u_m^{s_i},u_n^{p_j})=\frac{1}{n_1{n}_2}\underset{l=1}{\overset{n_1}{\&Sigma;}}\underset{f=1}{\overset{n_2}{\&Sigma;}}\frac{1}{2^{n_1-l+1}}sim(r^{s_i},r^{p_j})$

In formula, represent co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry similarity, n ₁and n ₂be respectively the senses of a dictionary entry with in attribute number, for the weighted value of the attribute of diverse location in senses of a dictionary entry definition, l is 1 to n ₁a variable, f is 1 to n ₂a variable, with for the senses of a dictionary entry with justice unit;

In formula, represent co-occurrence word s _iwith know net emotion word p _jjustice unit similarity, d is with path distance in hierarchical system, α is an adjustable parameter, and value is 1.6;

The degree of correlation of the co-occurrence word of Step5, calculating neologisms and neologisms:

$w_{ij}=\underset{r=0}{\overset{R}{\&Sigma;}}C(v_i,v_j)\&times;N(v_i,r,v_j)$

In formula, i and j represents the subscript of arbitrary two words, and R is the size of self-defining window, experimentally analyzes, and we get R=5, and r is the positive number being less than or equal to R, represents the distance of two words in R window, w _ijrepresent neologisms v _iwith neologisms v _ico-occurrence word v _jthe degree of correlation, N (v _i, r, v _j) be: neologisms v _iwith neologisms v _ico-occurrence word v _jco-occurrence number of times (r < R) when distance is r in R window in relevant documentation set, C (v _i, v _j)=R-r+1 is the co-occurrence intensity between two words;

Step6, with the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation in Step5, design of graphics model;

Step7, utilize Word similarity co-occurrence word s _ipolarity distribution on y:

$s_i\left(y\right)=\{\begin{array}{c}\frac{count\left(Sim\right(s_i,p_j^{+})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=+1\\ 0\\ \frac{count\left(Sim\right(s_i,p_j^{-})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=-1\end{array}y=0$

In formula, RK is threshold value, analyzes by experiment, and we get RK=0.25, s _iy () represents the polarity distribution of co-occurrence word, i and j represents the subscript of arbitrary two words, and Sim represents Words similarity, for knowing just tendentious word in net emotion word, for knowing in net emotion word negative tendentious word, the quantity of what count represented is word;

Step8, label propagation algorithm determination neologisms Sentiment orientation;

Step8.1 obtains the polarity distribution of neologisms, and its objective function is as follows:

In the polarity identification of neologisms, think that the degree of correlation between word is higher, the polarity distribution between them is more similar.Based on above thought, set up following objective function.Make objective function C minimum, and then obtain neologisms node v _ipolarity distribution.

$C=\underset{\underset{v_j\&Element;K\left(V_t\right)}{v_i\&Element;V_t}}{\&Sigma;}{w}_{ij}\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2+\mathrm{\&lambda;}\underset{v_i\&Element;V_t}{\&Sigma;}||q_i\left(t\right)-\mathrm{\&gamma;}|{|}^2$

Wherein:

$\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-s_i(y\left)\right)}^2$

$\left|\right|q_i\left(y\right)-\mathrm{\&gamma;}|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-\mathrm{\&gamma;}(y\left)\right)}^2$

In formula, i and j represents the subscript of arbitrary two words, q _iy () represents word node v _ipolarity distribution, s _iy () represents co-occurrence word v _jpolarity distribution, γ and λ is custom parameter, and value is 1 and 0.1, V respectively _trepresent co-occurrence word set, K (V _t) represent co-occurrence word V _tk nearest neighbor set of words, K value is 3 herein;

The distribution of the polarity of the neologisms obtained is designated as Q by Step8.2 _n, build linear classifier, obtain the Sentiment orientation of neologisms:

$Q_n=\left\{\begin{array}{cc}1& Q_n(y=1)-Q_n(y=-1)>RT\\ 0& |Q_n(y=1)-Q_n(y=-1)|\&le;RT\\ -1& Q_n(y=-1)-Q_n(y=1)>RT\end{array}\right.$

In formula, Q _n(y=1) Q is represented _nfor the probability of commendation, Q _n(y=-1) Q is represented _nfor the probability of derogatory sense, RT is threshold value, analyzes by experiment, and we get RT=0.6.As shown by the equation, when the probability that the Sentiment orientation of neologisms the is commendation probability deducted as derogatory sense is greater than threshold value RT, this neologisms Sentiment orientation is 1, is namely commendatory term; When the probability that the Sentiment orientation of neologisms the is commendation absolute value deducted as the probability of derogatory sense is less than threshold value RT, this neologisms Sentiment orientation is 0, is namely neutral words; When the probability that the Sentiment orientation of neologisms the is derogatory sense probability deducted as commendation is greater than threshold value RT, this neologisms Sentiment orientation is-1, is namely derogatory term.

Table 3 part neologisms

Polarity	Neologisms
		Commendation	To power, Bai Fumei, sprout god, milk tea younger sister, man god
Neutral	Private chat, little wind child, have drip, thief, do face, plunge into the commercial sea, cigarette friend
		Derogatory sense	To send out gram, overtax one's nerves, barrier set by the devil, brain deficiency, your younger sister, salty pig hand

Table 3 is depicted as the partial feeling neologisms obtained by experiment, and the emotion tendency of these neologisms is divided three classes, i.e. commendation, neutrality and derogatory sense.

By reference to the accompanying drawings the specific embodiment of the present invention is explained in detail above, but the present invention is not limited to above-mentioned embodiment, in the ken that those of ordinary skill in the art possess, various change can also be made under the prerequisite not departing from present inventive concept.

Claims (2)

1. the neologisms Sentiment orientation decision method based on microblogging, it is characterized in that: by Chinese word segmentation instrument, participle is carried out to microblogging language material, and with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, by word string combination of two adjacent in each piece, word string frequency after combination is added up, the word string of frequency higher than threshold value is gone here and there as neologisms candidate; According to the one-tenth word rule of Chinese and the rule of adjacent change number, filtration is carried out to neologisms candidate string and obtain neologisms; Recycling knows the sentiment dictionary of net, calculates co-occurrence word and the Words similarity knowing net emotion word; Calculate the degree of correlation of neologisms and co-occurrence word; With the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation of neologisms and co-occurrence word, design of graphics model; Utilize label propagation algorithm to obtain the feeling polarities distribution of neologisms, obtain the emotion tendency of neologisms finally by structure linear classifier.

2. the neologisms Sentiment orientation decision method based on microblogging according to claim 1, is characterized in that: the concrete steps of described method are as follows:

Step1, by Chinese word segmentation instrument, participle is carried out to microblogging language material;

Step2, with the stop words in word segmentation result for cut-point carries out piecemeal to the language material after participle, and by word string combination of two adjacent in each piece, the word string frequency after combination is added up, the word string of frequency higher than threshold value T is gone here and there as neologisms candidate;

Step3, rule that is regular according to the one-tenth word of Chinese and adjacent change number are carried out filtration to neologisms candidate string and are obtained neologisms;

Step4, utilize and know the sentiment dictionary of net, calculate co-occurrence word and the Words similarity knowing net emotion word:

Step4.1, find out the non-stop words with new Term co-occurrence in microblogging language material, as co-occurrence word;

Net sentiment dictionary is known in Step4.2, utilization, calculates co-occurrence word and the Words similarity knowing net emotion word, is expressed as follows:

In formula, Sim (s _i, p _j) represent co-occurrence word s _iwith know net emotion word p _jbetween Words similarity, i and j represents the subscript of arbitrary two words, m and n is respectively co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry number, represent co-occurrence word s _im the senses of a dictionary entry, represent and know net emotion word p _jn-th senses of a dictionary entry, P represents the set knowing net emotion word;

$\mathrm{si}m(u_m^{s_i},u_n^{p_j})=\frac{1}{n_1{n}_2}\underset{l=1}{\overset{n_1}{\&Sigma;}}\underset{f=1}{\overset{n_2}{\&Sigma;}}\frac{1}{2^{n_1-l+1}}\mathrm{sim}(r^{s_i},r^{p_j})$

In formula, represent co-occurrence word s _iwith know net emotion word p _jsenses of a dictionary entry similarity, n ₁and n ₂be respectively the senses of a dictionary entry with in attribute number, for the weighted value of the attribute of diverse location in senses of a dictionary entry definition, l is 1 to n ₁a variable, f is 1 to n ₂a variable, with for the senses of a dictionary entry with justice unit;

In formula, represent co-occurrence word s _iwith know net emotion word p _jjustice unit similarity, d is with path distance in hierarchical system, α is an adjustable parameter;

The degree of correlation of the co-occurrence word of Step5, calculating neologisms and neologisms:

$w_{ij}=\underset{r=0}{\overset{R}{\&Sigma;}}C(v_i,v_j)\&times;N(v_i,r,v_j)$

In formula, i and j represents the subscript of arbitrary two words, and R is the size of self-defining window, and r is the positive number being less than or equal to R, represents the distance of two words in R window, w _ijrepresent neologisms v _iwith neologisms v _ico-occurrence word v _jthe degree of correlation, N (v _i, r, v _j) be: neologisms v _iwith neologisms v _ico-occurrence word v _jco-occurrence number of times (r < R) when distance is r in R window in relevant documentation set, C (v _i, v _j)=R-r+1 is the co-occurrence intensity between two words;

Step6, with the co-occurrence word of neologisms and these neologisms for node, between neologisms and its co-occurrence word, set up limit, the weight being limit with the degree of correlation in Step5, design of graphics model;

Step7, utilize Word similarity co-occurrence word s _ipolarity distribution on y:

$s_i\left(y\right)=\{\begin{array}{c}\frac{count\left(Sim\right(s_i,p_j^{+})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=+1\\ 0\\ \frac{count\left(Sim\right(s_i,p_j^{-})>RK)}{count\left(Sim\right(s_i,p_j)>RK)}y=-1\end{array}y=0$

In formula, RK is threshold value, s _iy () represents the polarity distribution of co-occurrence word, i and j represents the subscript of arbitrary two words, and Sim represents Words similarity, for knowing just tendentious word in net emotion word, for knowing in net emotion word negative tendentious word, the quantity of what count represented is word;

Step8, label propagation algorithm determination neologisms Sentiment orientation;

Step8.1, obtain neologisms polarity distribution, its objective function is as follows:

$C=\underset{\underset{v_j\&Element;K\left(V_t\right)}{v_i\&Element;V_t}}{\&Sigma;}{w}_{ij}\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2+\mathrm{\&lambda;}\underset{v_i\&Element;V_t}{\&Sigma;}||q_i\left(t\right)-\mathrm{\&gamma;}|{|}^2$

Wherein:

$\left|\right|q_i\left(y\right)-s_i\left(y\right)|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-s_i(y\left)\right)}^2$

$\left|\right|q_i\left(y\right)-\mathrm{\&gamma;}|{|}^2=\underset{y}{\&Sigma;}{\left(q_i\right(y)-\mathrm{\&gamma;}(y\left)\right)}^2$

In formula, i and j represents the subscript of arbitrary two words, q _iy () represents neologisms node v _ipolarity distribution, s _iy () represents co-occurrence word v _jpolarity distribution, γ and λ is custom parameter, V _trepresent co-occurrence word set, K (V _t) represent co-occurrence word V _tk nearest neighbor set of words;

Step8.2, by the polarity of the neologisms obtained distribution be designated as Q _n, build linear classifier, obtain the Sentiment orientation of neologisms: when the probability that the Sentiment orientation of neologisms the is commendation probability deducted as derogatory sense is greater than threshold value RT, this neologisms Sentiment orientation is 1, is namely commendatory term; When the probability that the Sentiment orientation of neologisms the is commendation absolute value deducted as the probability of derogatory sense is less than threshold value RT, this neologisms Sentiment orientation is 0, is namely neutral words; When the probability that the Sentiment orientation of neologisms the is derogatory sense probability deducted as commendation is greater than threshold value RT, this neologisms Sentiment orientation is-1, is namely derogatory term;

$Q_n=\left\{\begin{array}{cc}1& Q_n(y=1)-Q_n(y=-1)>RT\\ 0& |Q_n(y=1)-Q_n(y=-1)|\&le;RT\\ -1& Q_n(y=-1)-Q_n(y=1)>RT\end{array}\right.$

In formula, Q _n(y=1) Q is represented _nfor the probability of commendation, Q _n(y=-1) Q is represented _nfor the probability of derogatory sense, RT is threshold value.