CN108388608A

CN108388608A - Emotion feedback method, device, computer equipment and storage medium based on text perception

Info

Publication number: CN108388608A
Application number: CN201810119475.0A
Authority: CN
Inventors: 黄译萱; 蒋菲; 陈桓; 张良杰
Original assignee: Kingdee Software China Co Ltd
Current assignee: Kingdee Software China Co Ltd
Priority date: 2018-02-06
Filing date: 2018-02-06
Publication date: 2018-08-10
Anticipated expiration: 2038-02-06
Also published as: CN108388608B

Abstract

This application involves a kind of emotion feedback method, device, computer equipment and storage mediums based on text perception.This method includes：The first weight distribution ratio and the second weight distribution ratio are calculated to the content of text to be feedback got by content of text perceptron；Content of text to be feedback is inputted to the neural network trained, obtains emotional semantic classification result；Corresponding affection index result is calculated according to preset rules in dictionary are preset in emotion word radix in content of text to be feedback；Emotional semantic classification result is compared according to the first weight distribution ratio, the second weight distribution and affection index result is weighted, and obtains sentiment analysis result；Obtain the corresponding emotion viewpoint result of content of text to be feedback；It is replied from default comment according to sentiment analysis result and emotion viewpoint result and chooses target emotion feedback comments reply corresponding with content of text to be feedback in correlation database.The matching rate that content of text to be feedback is replied with target emotion feedback can be improved using this method.

Description

Emotion feedback method, device, computer equipment and storage medium based on text perception

Technical field

This application involves field of computer technology, more particularly to a kind of emotion feedback method, dress based on text perception It sets, computer equipment and storage medium.

Background technology

Affective interaction generally refer to the content source of analysis such as document etc with determine conveyed by content source it is specific anti- It answers or attitude, then carries out intelligent entertaining emotion feedback on the basis of sentiment analysis contacts.If for example, surfing the Internet in Taobao's client Shopping product feel dissatisfied, then difference have been given to comment, and computer can be given and be fed back with interesting feelings according to sentiment analysis result automatically at this time The case where reply of sense, such as parent, you say, is implicitly present in us and is continuously improving this appearance, woulds you please happily spend every day ！

Currently, the common text emotion trend analysis of mainstream, which has, classifies based on neural network characteristics or is based on sentiment dictionary Calculating emotion score, although however the emotion text content that is sorted in chapter paragraph level based on neural network characteristics differentiates upper have Very big advantage and accuracy, but cannot obtain that chapter paragraph is carried out specifically to score and extract emotion word, and it is based on emotion Although the algorithm of dictionary can accurately extract emotion word and there can be emotional semantic classification very much, instructed in the long text of chapter paragraph White silk is ineffective.Therefore, the emotion of the emotion text content of chapter paragraph level feed back, traditional technology based on neural network spy Sign classification or all None- identifieds of the algorithm based on sentiment dictionary, identification matching rate is low, ineffective.

Invention content

Based on this, it is necessary in view of the above technical problems, provide a kind of that can improve text and corresponding feedback content Emotion feedback method, device, computer equipment and the storage medium based on text perception with rate.

A kind of emotion feedback method based on text perception, this method include：

Obtain content of text to be feedback；

Feedback content of text is treated by content of text perceptron, and the first weight distribution ratio and the second weight point is calculated Proportioning；

Content of text to be feedback is inputted to the neural network trained, obtains emotional semantic classification result；

The content of text to be feedback is split, multiple emotion words in the content of text to be feedback are obtained；

Corresponding affection index result is calculated according to preset rules in dictionary are preset in multiple emotion word radix；

Emotional semantic classification result is compared according to the first weight distribution ratio, the second weight distribution and affection index result is added Power, obtains the sentiment analysis result of content of text to be feedback；

Obtain the corresponding emotion viewpoint result of content of text to be feedback；

It is replied in correlation database and is chosen and text to be feedback from default comment according to sentiment analysis result and emotion viewpoint result The corresponding target emotion feedback comments of content are replied.

A kind of emotion feedback device based on text perception, the device include：

Content of text acquisition module to be feedback, for obtaining content of text to be feedback；

Weight distribution treats feedback content of text than computing module, for passing through content of text perceptron and is calculated first Weight distribution ratio and the second weight distribution ratio；

Emotional semantic classification result acquisition module obtains feelings for content of text to be feedback to be inputted the neural network trained Feel classification results；

Affection index result computing module is split for treating feedback content of text, obtains content of text to be feedback In multiple emotion words, by multiple emotion word radix according to preset dictionary in preset rules corresponding affection index knot is calculated Fruit；

Sentiment analysis interpretation of result module, for comparing emotional semantic classification according to the first weight distribution ratio, the second weight distribution As a result it is weighted with affection index result, obtains the sentiment analysis result of content of text to be feedback；

Emotion viewpoint result acquisition module, for obtaining the corresponding emotion viewpoint result of content of text to be feedback；

Feedback module is chosen for being replied in correlation database from default comment according to sentiment analysis result and emotion viewpoint result Target emotion feedback comments corresponding with content of text to be feedback are replied.

A kind of computer equipment, including memory, processor and storage can be run on a memory and on a processor Computer program, the processor realize following steps when executing the computer program：

Obtain content of text to be feedback；

It treats feedback content of text to be split, obtains multiple emotion words in content of text to be feedback；

A kind of computer readable storage medium, is stored thereon with computer program, and the computer program is held by processor Following steps are realized when row：

Obtain content of text to be feedback；

Above-mentioned emotion feedback method, device, computer equipment and storage medium based on text perception, pass through content of text Perceptron obtains the first weight distribution ratio and the second weight distribution ratio to the content of text intelligence computation to be feedback got.It will obtain The neural network trained of content of text to be feedback input got obtains corresponding emotional semantic classification as a result, and will get Corresponding affection index knot is calculated according to preset rules in dictionary are preset in multiple emotion word radix in content of text to be feedback Fruit.Feedback content of text is treated according to content of text perceptron, and the first weight distribution ratio and the comparison of the second weight distribution is calculated Emotion allocation result and affection index result are weighted, and obtain the sentiment analysis result of content of text to be feedback.Further obtain The corresponding emotion viewpoint of content of text to be feedback is taken as a result, sentiment analysis result and emotion viewpoint result are tied as a whole Structure data go default comment to reply and match corresponding target emotion comment reply in correlation database, during the comment of target emotion is replied The feedback content for replying as content of text to be feedback.Therefore, by content of text perceptron by the output knot of neural network The arithmetic result of fruit and dictionary is combined, finally automatically derive target emotion feedback reply, improve content of text to be feedback with The matching rate that target emotion feedback is replied.

Description of the drawings

Fig. 1 is the flow diagram of the emotion feedback method based on text perception in one embodiment；

Fig. 2 is to treat feedback content of text in one embodiment to be split, and is obtained multiple in content of text to be feedback Multiple emotion word radix evidence is preset the step of corresponding affection index result is calculated in preset rules in dictionary by emotion word Flow diagram；

Fig. 3 is flow signal the step of obtaining the emotion viewpoint result of content of text to be feedback answered in one embodiment Figure；

Fig. 4 is the flow diagram for the generation step that cluster viewpoint library is preset in one embodiment；

Fig. 5 is the flow diagram of the generation step of neural network in one embodiment；

Fig. 6 is the flow diagram for the generation step that correlation database is replied in comment in one embodiment；

Fig. 7 is the flow diagram of the emotion feedback method based on text perception in another embodiment；

Fig. 8 is the principle schematic of the emotion feedback method based on text perception in one embodiment；

Fig. 9 is the feedback result schematic diagram of the emotion feedback method based on text perception in one embodiment；

Figure 10 is the structure diagram of the emotion feedback device based on text perception in one embodiment；

Figure 11 is the structure diagram of emotion viewpoint result acquisition module in one embodiment；

Figure 12 is the internal structure chart of one embodiment Computer equipment.

Specific implementation mode

It is with reference to the accompanying drawings and embodiments, right in order to make the object, technical solution and advantage of the application be more clearly understood The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not For limiting the application.

In one embodiment, as shown in Figure 1, a kind of emotion feedback method perceived based on text is provided, with the party Method is applied to illustrate for server or terminal, includes the following steps：

Step 102, content of text to be feedback is obtained.

Wherein, content of text to be feedback here, including but not limited to various articles, comment etc..Specifically, can pass through The relevant application program of terminal uploads content of text, using the content of text of upload as content of text to be feedback.Here application Program can be but not limited to the various news applications that can deliver content of text to be feedback, Video Applications, social networking application, Forum's application etc..

Step 104, feedback content of text is treated by content of text perceptron and the first weight distribution ratio and the is calculated Two weight distribution ratios.

Wherein, content of text perceptron is the two classification device algorithm for linear separability data set, it may also be said to be to adopt The brain sensing capability transmitted with simulation multi-neuron simulates automatic identification simulator of the mankind for thinking judgment model.It is logical Content of text perceptron is crossed the content of text to be feedback that terminal is got to be carried out the first weight distribution ratio and second is calculated Weight distribution ratio.So-called weight distribution ratio be by the relative importance to one of index in the overall evaluation, can be according to According to the more principle of the weight of weight distribution corresponding distribution more bigger than numerical value.Specifically, content of text to be feedback is being got Afterwards, content of text perceptron can calculate the content of text to be feedback got, respectively obtain the first weight distribution ratio and Second weight distribution ratio.

Step 106, content of text to be feedback is inputted to the neural network trained, obtains emotional semantic classification result.

Wherein, neural network here is artificial neural network, is a kind of imitation animal nerve network behavior feature, is carried out The algorithm mathematics model of distributed parallel information processing.Since trained neural network can automatically divide the content of input Class, therefore after getting content of text to be feedback, content of text to be feedback is input to trained neural network, is obtained Classification results corresponding with the input content of neural network, i.e., output result corresponding with content of text to be feedback are emotional semantic classification As a result.Wherein, emotional semantic classification result has preset fractional value.So-called emotional semantic classification is the result is that by will be in text to be feedback Appearance is input to the output that trained neural network has obtained as a result, including but not limited to actively, passiveness, praising, like.

Step 108, it treats feedback content of text to be split, obtains multiple emotion words in content of text to be feedback.

Step 110, multiple emotion word radix in content of text to be feedback are calculated according to preset rules in dictionary are preset To corresponding affection index result.

Wherein, it is the word being obtained ahead of time in the comment content that manual examination and verification mark out to here preset at dictionary, due to comment The sentence that content is made of multiple words, therefore comment content is split to obtain multiple words, it will be multiple after segmentation Word carries out cleaning and handles to obtain final dictionary.Therefore after getting content of text to be feedback, due in text to be feedback The emotion word of emotion can be represented by having in appearance, then content of text to be feedback is needed to be split, and be obtained in content of text to be feedback Multiple emotion words.Further, multiple emotion word radix in content of text to be feedback are default in dictionary according to presetting Corresponding affection index result is calculated in rule.So-called preset rules are the weight rule of preset emotion word in dictionary Then.Wherein, emotion word here includes but not limited to the emotion word etc. indicated actively, passive, happy, sad.Specifically, After getting content of text to be feedback, according to the weight rule of the preset emotion word in default dictionary to getting Content of text to be feedback in the weight calculations of multiple emotion words obtain emotion score corresponding with content of text to be feedback, That is affection index result.

Step 112, emotional semantic classification result and affection index result are compared according to the first weight distribution ratio, the second weight distribution It is weighted, obtains the sentiment analysis result of content of text to be feedback.

Wherein, sentiment analysis emotional semantic classification result and affection index result the result is that be made of.Specifically, pass through text Perception of content device treats feedback content of text and carries out that the first weight distribution ratio and the second weight distribution ratio is calculated, according to first Weight distribution is than comparing the corresponding fractional value of emotional semantic classification result of neural network output or default word with the second weight distribution The affection index result that allusion quotation is calculated is weighted, and obtains the corresponding sentiment analysis result of content of text to be feedback.Institute It is to assign different weights to the corresponding fractional value of emotional semantic classification result and affection index result to calculate to call weighted calculation.

Step 114, the corresponding emotion viewpoint result of content of text to be feedback is obtained.

Wherein, emotion viewpoint according to stating the fact by content of text the result is that be summarised as a total viewpoint, such as " can will send Goods is fast, logistics rapidly, pass power soon " etc. same class describe the rapid content of text of logistics and be classified as " logistics is rapidly ", i.e. " logistics Rapidly " as " delivery is fast, logistics rapidly, pass power soon " etc. same class description viewpoint.Specifically, according in text to be feedback Hold the content obtaining emotion viewpoint result corresponding with content of text to be feedback of statement.

Step 116, it is replied in correlation database from default comment according to sentiment analysis result and emotion viewpoint result and chooses and wait for The corresponding target emotion feedback comments of content of text are fed back to reply.

Wherein, it is that the comment that manual examination and verification mark out will be obtained ahead of time to reply sample that correlation database is replied in default comment here It is associated with corresponding sample sentiment analysis result and sample viewpoint result, correlation database, wherein sample are replied in obtained comment It includes comment sample and corresponding reply sample that comment, which is replied,.For example, comment sample is：" logistics rapidly, favorable comment ", sample return It is again：" thank you ", sample sentiment analysis result corresponding with comment reply sample are：" like praise, 10 points ", and is commented on back The corresponding sample viewpoint result of duplicate sample sheet is：" logistics is rapidly ", sample is replied in comment is：It " logistics rapidly, favorable comment ", " thanks You " it is " like praise, 10 points " with sample sentiment analysis result and sample viewpoint result is to be associated " logistics is rapidly ", it obtains Correlation database is replied to comment.Specifically, after getting sentiment analysis result and emotion viewpoint result, by sentiment analysis result and Emotion viewpoint result is replied in correlation database from default comment according to structural data and is chosen and structuring number as structural data Sample is replied according to corresponding comment, comment is replied into the reply sample in sample as the corresponding target feelings of content of text to be feedback Feel feedback comments to reply.

In the above-mentioned emotion feedback method based on text perception, by content of text perceptron to the text to be feedback that gets The intelligence computation of this content obtains the first weight distribution ratio and the second weight distribution ratio.The content of text to be feedback got is inputted The neural network trained obtains corresponding emotional semantic classification as a result, and by multiple feelings in the content of text to be feedback got Corresponding affection index result is calculated according to preset rules in default dictionary in sense word.It is treated according to content of text perceptron The first weight distribution ratio is calculated in feedback content of text and the second weight distribution compares emotion allocation result and affection index knot Fruit is weighted, and obtains the sentiment analysis result of content of text to be feedback.Further obtain the corresponding feelings of content of text to be feedback Sense organ point is as a result, structural data replys correlation database from default comment as a whole by sentiment analysis result and emotion viewpoint result It is middle to match corresponding target emotion comment reply, reply as the anti-of content of text to be feedback during the comment of target emotion is replied Present content.Therefore, the arithmetic result of the output result of neural network and dictionary is combined by content of text perceptron, most Target emotion feedback is automatically derived eventually to reply, and improves the matching rate that content of text to be feedback is replied with target emotion feedback.

In one embodiment, it is split as shown in Fig. 2, treating feedback content of text, obtains content of text to be feedback In multiple emotion words, by multiple emotion word radix according to preset dictionary in preset rules corresponding affection index knot is calculated Fruit, including：

Step 202, it treats feedback content of text to be split, obtains multiple sentences.

Since the content of text to be feedback got is usually comment or the article in the form of paragraph, need to treat anti- Feedback content of text is split, multiple sentences after being divided.It specifically, can be right after getting content of text to be feedback Content of text to be feedback is split according to certain rule, obtains multiple sentences.Wherein, segmentation rule is point with punctuation mark Mark is cut, single content of text to be feedback is divided into N number of sentence.

Step 204, each sentence is weighted according to special word, obtains the sentence after multiple weightings.

Wherein, special word includes but not limited to adversative conjunction etc., since special word can represent the degree of emotion, The sentence at place can be weighted according to special word.So-called weighting is to confer to the weight degree of the sentence where special word Larger.Specifically, after the content of text to be feedback to acquisition is split to obtain multiple sentences, according in each sentence The weight that special word assigns the sentence where special word is more higher, i.e. the higher degree for representing emotion of weight is bigger.For example, Content of text to be feedback, which is divided into multiple sentences, is：" this computer appearance is pretty good, but has used two days, there is a power consumption ！", due to being " still " to belong to adversative conjunction, i.e., special word, then assign " but two days have been used, there is a power consumption！" this The weight of a sentence is 0.6, and the weight of " this computer appearance is pretty good " is 0.4.

Step 206, each sentence is split, obtains multiple emotion words.

Since multiple sentences after segmentation are with the comment of sentential form or article, it is therefore desirable to multiple after segmentation Sentence is split, multiple words after being divided.Specifically, treat feedback content of text be split to obtain multiple sentences After son, each sentence can be split according to certain rule, obtain multiple words, further extracted from multiple words To multiple emotion words.Wherein, segmentation rule can be according to ingredient of the word in sentence, the collocation of the part of speech of word and word Custom is split.

Step 208, multiple emotion words are calculated according to the weighted score of each emotion word in default dictionary to correspond to Emotion word score.

Specifically, after the emotion word in extracting each sentence, according to assigning each emotion word in default dictionary The corresponding emotion word score of multiple emotion words is calculated in weighted score.Such as：" this computer appearance is pretty good, still With two days, there is a power consumption！" wherein, " good " and " power consumption " is respectively as positive emotion word and passive emotion Word, the weight due to presetting imparting " good " in dictionary is 2 points, and the weight for assigning " power consumption " is -1 point, is further able to count The emotion score for respectively obtaining " good " and " power consumption " is respectively " 2 points " and " -1 point ".

Step 210, the default word adjacent with emotion word is searched according to each emotion word.

Step 212, each sentence is calculated with emotion word score according to the weighted score of default word in default dictionary The corresponding emotion score of son.

Step 214, the affection index result of content of text to be feedback is calculated according to the emotion score of each sentence.

Wherein, default word is degree adverb, negative word etc. to be found successively forward, by degree on the basis of emotion word Adverbial word, negative word are as default word.Further, it is searched and is sequentially looked forward and emotion word on the basis of each emotion word The adjacent degree adverb of language, negative word etc..After finding the default words such as degree adverb, negative word, according in default dictionary Score value corresponding with default word is calculated in weighted score to presetting word.Further, it is being calculated and default word After the corresponding score value of language, the corresponding emotion point of each sentence is calculated in the score value of default word and the score of emotion word Number, the affection index result of the content of text to be feedback got is calculated further according to the corresponding emotion score of each sentence.

Such as：Content of text to be feedback is：" this computer appearance is pretty good, but has used two days, there is a power consumption ！", content of text to be feedback is divided into multiple sentences with comma, by being split to multiple sentences, extracts each sentence In emotion word.From default dictionary matching obtain representing positive emotion word as：" good " presets dictionary to the emotion The score of word is set as 1 point.Further, it is pre- that degree adverb or negative word etc. are found forward successively on the basis of " good " If word, it is " suitable " that lookup, which obtains degree adverb, is 2 fractions to the score that degree adverb is " suitable " according to default dictionary The fractional value of emotion word is multiplied the fractional value of sentence where obtaining emotion word, i.e. " phase by weight with the fractional value of degree adverb When good " fractional value of place " this computer appearance is pretty good " is 2 points.

Since " still " is adversative conjunction, it is 0.6 to preset and assign the weight of " still " in dictionary, wherein " but use Two days, there is a power consumption！" in sentence, find and represent passive emotion word " consumption ", disappear to the expression wherein presetting in dictionary The score of " consumption " of pole is set as -1 point.It is " a little " due to finding degree adverb on the basis of " consumption ".Due to presetting dictionary 1.25 points, therefore the fractional value for " having a power consumption " behind " still " are set as to the fractional value that degree adverb is " a little " It is -1.25, simultaneously because " still " this adversative conjunction, subsequent sentence is only that entire content of text to be feedback is most important thinks table The emotion reached is to indicate that the emotion degree for the sentence praised is more important than front half section.So multiplied by sentence before and after " still " Different weights are added again, are obtained the last fractional value of content of text to be feedback and are：2*0.4+-1.25*0.6, i.e., in text to be feedback The affection index result of appearance.

In one embodiment, as shown in figure 3, obtaining the emotion viewpoint of content of text to be feedback answered as a result, including：

Step 302, it treats feedback content of text and carries out syntactic type analysis, obtain the corresponding syntax of content of text to be feedback Type.

Wherein, syntactic type refers to combining to constitute according to certain rule between word and word.And text to be feedback The sentence that content is made of multiple words, in a practical situation, due to there are many combinations between word, then multiple words Language combine the sentence to be formed also have it is multiple.In this case, it is necessary to treat feedback content of text and carry out syntactic type point Analysis.When carrying out syntactic type analysis, can content of text to be feedback be first divided into multiple words, then to multiple words after segmentation Language carries out syntactic type analysis, obtains corresponding syntactic type, wherein syntactic type analysis refers to which determining word constitutes one A phrase, which word are verb, adjective or adverbial word etc..Such as：Content of text to be feedback is：" computer screen is a little Greatly ", since " computer screen " is noun n, " big " is adjective a, and " a little " is adverbial word d, then treats feedback content of text and carry out The syntactic type that syntactic type analyzes to obtain content of text to be feedback is：“n+d+a”.

Step 304, when syntactic type in default cluster viewpoint library when the corresponding viewpoint syntactic type of viewpoint, according to word to The corresponding matching syntactic type of syntactic type is calculated in amount model.

Wherein, it refers to same or similar previously according to same class content of text states the fact to preset cluster viewpoint library It is summarised as a total viewpoint, obtains cluster viewpoint library.Due to viewpoint be typically also in the form of sentence present, and sentence be by Multiple word compositions, and between word and word composition, i.e. viewpoint syntactic type are combined according to certain rule.Therefore, exist After obtaining the syntactic type of content of text to be feedback, when the syntactic type of content of text to be feedback detects in default viewpoint library When the corresponding viewpoint syntactic type of viewpoint, the syntactic type and viewpoint syntax of content of text to be feedback are calculated according to term vector model The similarity of type then can be using the viewpoint syntactic type as text to be feedback when similarity is more than default similarity threshold The matching syntactic type of the syntactic type of content.Here term vector model is for calculating the syntactic type of content of text and gathering The similarity of viewpoint syntactic type in class viewpoint library.

Step 306, according to matching syntactic type match from viewpoint syntactic type obtain with match syntactic type it is corresponding Emotion viewpoint result.

Specifically, since viewpoint is presented in the form of sentence, sentence is made of multiple words, and word and word it Between constitute according to certain rule combination, i.e. viewpoint syntactic type.That is each viewpoint has corresponding viewpoint syntax Type.After getting the matching syntactic type of content of text to be feedback, according to matching syntactic type from cluster viewpoint library It is matched in the corresponding viewpoint syntactic type of viewpoint.When successful match, the corresponding viewpoint of syntactic type will be matched as waiting for Feed back the emotion viewpoint result of content of text.

In one embodiment, as shown in figure 4, the generation step in default cluster viewpoint library includes：

Step 402, multiple comment samples are split, obtain the corresponding multiple words of each comment sample.

Wherein, comment sample is several macrotaxonomies (hotel, the hand that major electric business platform is acquired using distributed reptile technology Machine, household electrical appliances, books etc.) commodity it is each up to ten thousand comment sample.So-called distributed reptile exactly disposes tens of in different geographical Server creates hundreds of containers using Docker lightweight virtual machine technologies on every server, is filled on each container Reptile module is clicked in the webpage simulation for carrying Java exploitations, forms large-scale distributed reptile network.One large-scale reptile task is drawn It is divided into a large amount of small-sized subtask and builds task queue, the Task Scheduling Mechanism for then leading to more container collaborations too much appoints son It is executed in business distribution according to need to each container, to realize the concurrent distributed reptile of superelevation.Specifically, since comment sample is The sentence being made of multiple words, therefore comment sample is split to obtain multiple words, wherein segmentation sentence can be according to word Ingredient, the part of speech of word and the collocation custom of word of the language in sentence are split.

Step 404, multiple words corresponding to each comment sample carry out part-of-speech tagging, obtain each comment sample and correspond to Syntactic type.

Specifically, multiple to what is obtained after segmentation after the word for being split to obtain after multiple segmentations to comment sample Word carries out part-of-speech tagging, is ranked up, is obtained to multiple words after mark in the position where sentence according to the word of mark To the corresponding syntactic type of comment sample being made of multiple words.Such as：It is " computer screen is somewhat big " to comment on sample, to comment Sample is split to obtain multiple words：" computer screen ", " a little ", " big ".Since " computer screen " is noun n, " big " For adjective a, " a little " is adverbial word d, then is analyzed for " computer screen is somewhat big " progress syntactic type comment sample and commented It is by the syntactic type that sample is " computer screen is somewhat big "：“n+d+a”.

Step 406, the phase of the corresponding syntactic type of each comment sample is calculated according to trained term vector model Like degree.

Step 408, when similarity is more than default similarity, the syntactic type that similarity is more than to default similarity corresponds to Multiple comment samples merge, obtain comment sample set.

Specifically, trained term vector model here is for calculating between the syntactic type of each content of text Similarity.After the syntactic type for getting each comment sample, each comment is calculated according to trained term vector model The similarity of the corresponding syntactic type of sample.Wherein similarity is the similar journey between the corresponding syntactic type of each comment sample Degree, similarity degree is higher to indicate more similar.Further, the similarity of the corresponding syntactic type of each comment sample is being obtained Afterwards, when similarity is more than default similarity, similarity is more than to the corresponding multiple comment samples of syntactic type of default similarity Originally it merges, obtains comment sample set.Such as：Commenting on sample is：" logistics is rapidly ", " delivery is fast ", " passing power soon ", according to The similarity of the corresponding syntactic type of each comment sample is calculated in trained term vector model, since comment sample is： " logistics is rapidly ", " delivery is fast ", " passing power soon " the similarity of syntactic type be more than default similarity, then will " logistics god Speed ", " delivery is fast ", " passing power soon " are as a comment sample set.

Step 410, centre word is determined as commenting on the viewpoint of sample set by the centre word for obtaining comment sample set, will Viewpoint is added in default cluster viewpoint library, generates default cluster viewpoint library, and wherein centre word is used to define the type of comment sample.

Specifically, since the element in comment sample set belongs to the statement same class fact, comment sample is being obtained After this set, obtain can summarize or Summarizing comment sample set in element it is stated the fact centre word, and will obtain The centre word got is determined as commenting on the viewpoint of sample set, which is added into default cluster viewpoint library, generates default Cluster viewpoint library.Wherein, centre word is used to summarize or summarize the word for the fact that the element in comment sample set is stated Or sentence, that is, define the type for the comment sample set being made of of a sort multiple comment samples.Such as：Comment on sample Element is in set：" logistics is rapidly ", " delivery is fast ", " passing power soon ", since comment sample set element belongs to same class The comment sample of logistics rapidly is described, therefore the centre word for obtaining comment sample set is：" logistics is rapid ", by " logistics is fast Viewpoint of the speed " as the comment sample set, further " logistics is rapid " is added in viewpoint library.It then presets in cluster viewpoint library " logistics is rapid " viewpoint is to be for describing comment sample set：" logistics is rapidly ", " delivery is fast ", " passing power soon ".

In one embodiment, as shown in figure 5, the generation step of neural network includes：

Step 502, the comment sample marked is obtained.

Step 504, comment sample is inputted into trained term vector analysis model word2Vec, obtains comment sample and corresponds to Term vector.

Step 506, it is inputted term vector as feature in neural network, neural network is trained, is trained Neural network.

Specifically, since comment sample is a few macrotaxonomy commodity for acquiring major electric business platform using distributed reptile technology It is each up to ten thousand comment sample, then the comment sample to crawling carry out part-of-speech tagging, obtain the comment sample marked.Into one Step ground, will get the comment sample marked and is input to trained term vector analysis model word2Vec, commented on The corresponding term vector of sample.Wherein, term vector analysis model word2Vec is for switching to calculate the words in natural language The neural network for the dense vector that machine is appreciated that.Dense vector can be described as term vector again.So-called term vector represents word It is semantic, can be used for doing and classify, cluster and the expression symbol of similarity calculation.Obtaining the corresponding term vector of comment sample Afterwards, it is input to term vector as feature in neural network, unsupervised training is carried out to neural network and Training obtains To trained neural network.

In one embodiment, as shown in fig. 6, the generation step that correlation database is replied in comment includes：

Step 602, it obtains the comment that has marked and replys sample, wherein it includes comment sample and corresponding that sample is replied in comment Reply sample.

Specifically, it is several macrotaxonomies that major electric business platform is acquired using distributed reptile technology to reply sample due to comment Sample is replied in each up to ten thousand comments of commodity, then the comment to crawling replys sample and carries out part-of-speech tagging, obtains having marked Sample is replied in comment.It includes comment sample and corresponding reply sample that wherein sample is replied in comment, will comment sample and right The reply sample answered replys sample as comment.

Step 604, by content of text perceptron to comment reply sample be calculated first sample weight distribution ratio and Second sample weights distribution ratio.

Specifically, since content of text perceptron is that the brain sensing capability transmitted using simulation multi-neuron simulates the mankind It is a kind of mechanism for judgement for thinking deeply the automatic identification simulator of judgment model.Therefore, it has marked getting After sample is replied in comment, content of text perceptron can be calculated getting comment reply sample, respectively obtain first sample Weight distribution ratio and the second sample weights distribution ratio.Wherein, weight distribution ratio be by one of index in the overall evaluation In relative importance, weight distribution is bigger than more high corresponding weight.

Step 606, sample input trained neural network is replied into comment, obtains sample emotional semantic classification result.

Specifically, it since trained neural network can automatically classify to the content of input, gets and comments After replying sample, sample is replied into comment and is input in trained neural network, is obtained in the input with neural network Hold corresponding classification results, i.e., output result corresponding with comment reply sample is emotional semantic classification result.Wherein, emotional semantic classification knot Fruit include but not limited to actively, it is passive, the types such as praise, like, criticizing.Such as：Sample is replied in comment：" oh！Logistics god Through this screen is good-looking, and mainly its price is low " it is comment sample, " so high praise do other is too shy " is to reply sample Comment reply sample is input in trained neural network by this, and trained neural network can be in input Appearance is classified, and is obtained the corresponding emotional semantic classification result of comment reply sample and is：" like, praise ".Further, it obtains pre- It is to emotional semantic classification result first：The score that " like, praise " is set as：10 points.

Step 608, multiple emotion word radix in sample are replied in comment to be calculated according to preset rules in dictionary are preset Sample affection index result.

Specifically, it includes multiple emotion words to be replied in sample due to comment, after getting comment and replying sample, Multiple feelings according to the weight rule of the preset emotion word in default dictionary to the content of text to be feedback got The weight calculation of sense word obtains emotion score corresponding with content of text to be feedback, i.e. affection index result.Such as：Comment is replied Sample is：" oh！Rapidly, this screen is good-looking for logistics, and mainly its price is low " it is comment sample, " so people is done to obtain in high praise Family is too shy " it is to reply sample, the emotion word in extraction comment sample is：" good-looking ", since default dictionary is to the emotion The score of word is set as 2 points, and the weight of the sentence where the emotion word is 0.6, then the emotion of sample is replied in the comment Index results are：2*0.6.

Step 610, according to first sample weight distribution ratio, the second sample weights distribution ratio to sample emotional semantic classification result and Sample affection index result is weighted, and obtains the sample sentiment analysis result that sample is replied in comment.

Specifically, sample sentiment analysis is sample emotional semantic classification result and sample affection index result the result is that be made of, Therefore sample is replied to comment by content of text perceptron to carry out that first sample weight distribution ratio and the second sample is calculated Weight distribution ratio corresponds to sample emotional semantic classification result according to first sample weight distribution ratio, the second sample weights distribution ratio and divides Numerical value and sample affection index result are weighted, and obtain the sample sentiment analysis result that sample is replied in comment.Such as：Pass through It is 20% that content of text perception replys the first sample weight distribution ratio that sample is calculated to comment, the second sample weights Distribution ratio is 80%, and first sample weight distribution score corresponding with the sample emotional semantic classification result exported based on neural network The sample affection index result that value is multiplied and the second sample weights distribution ratio is calculated with default dictionary carries out phase Multiply, finally obtains the sample sentiment analysis result that sample is replied in comment.

Step 612, it obtains comment and replys the corresponding sample emotion viewpoint result of sample.

Step 614, sample sentiment analysis result and emotion viewpoint result are associated to obtain with that will comment on reply sample Sample is replied in comment after association, and replying sample according to the comment after association generates comment reply correlation database.

Specifically, according to get comment reply sample statement the corresponding sample emotion viewpoint of content obtaining as a result, The sample emotion viewpoint result got is replied sample with comment as a whole with sample sentiment analysis result to close Sample is replied in connection, the comment after being associated with.Further, sample is replied into the comment after association and comment reply correlation database is added In.Such as：Commenting on the comment sample replied in sample is：" your this product interaction hell to pay ", replying sample is：It " obeys the command！ Small horse back goes to make into ", sample sentiment analysis result, which is calculated, is：" passive, -5 points ", gets sample emotion viewpoint result For：Therefore comment sample is by " product interaction trouble "：" your this product interaction hell to pay ", replying sample is：It " abides by Life！Small horse back goes to make into ", sample sentiment analysis result is：" passive, -5 points " and sample emotion viewpoint result are：" production After product interaction trouble " is associated, the overall data after being associated with is added to comment and replys in correlation database.

Fig. 7 shows the flow chart of the emotion feedback method perceived based on text in one embodiment, and this method is to be applied to It illustrates, includes the following steps for server or terminal：

Step 702, content of text to be feedback is obtained.

Step 704, feedback content of text is treated by content of text perceptron and the first weight distribution ratio and the is calculated Two weight distribution ratios.

As shown in figure 8, Fig. 8 shows the principle schematic of the emotion feedback method perceived based on text in one embodiment. After getting content of text to be feedback, feedback content of text is treated by content of text perceptron and carries out weight calculation, respectively Obtain the first weight distribution ratio and the second weight distribution ratio.Wherein, the first weight distribution is divided than corresponding based on neural network emotion The emotional semantic classification of analysis as a result, the sentiment analysis of the second weight distribution allusion quotation more word-based than correspondence affection index result.If the first power It reassigns than the second weight distribution than big, then the weight bigger that the sentiment analysis result based on neural network sentiment analysis occupies, Conversely, the weight bigger that then the affection index result of the sentiment analysis based on dictionary occupies.

Step 706, content of text to be feedback is inputted to the neural network trained, obtains emotional semantic classification result.

Since trained neural network can automatically classify to the content of input, as shown in figure 8, being waited for instead getting After presenting content of text, content of text to be feedback is input in trained neural network, the input with neural network is obtained Content is the corresponding classification results of content of text to be feedback, i.e. emotional semantic classification result.Such as：The content of text to be feedback got For：" oh, this logistics is very fast ", then the content of text to be feedback is input in trained neural network obtain it is corresponding Emotional semantic classification result is：" like, praise ".Further, obtaining preset emotional semantic classification result is：" like, praise " is right The fractional value answered.

Step 708, it treats feedback content of text to be split, obtains multiple sentences.

Step 710, each sentence is weighted according to special word, obtains the sentence after multiple weightings.

Step 712, each sentence is split, obtains multiple emotion words.

Specifically, since content of text to be feedback is usually comment or article in the form of paragraph, it is therefore desirable to treat It is that segmentation identifies to feed back content of text with punctuation mark, multiple sentences that content of text to be feedback is divided.To obtaining To content of text to be feedback be split after obtaining multiple sentences, can be according to special words such as adversative conjunctions in each sentence The weight that more high point is assigned to the sentence where special word, obtains the sentence after each weighting.Due to the sentence after each weighting Son be with the comment of sentential form or article, therefore can be according to ingredient of the word in sentence, the part of speech of word and word Collocation custom the sentence after weighting is split, multiple words after being divided, further from multiple after segmentation Multiple emotion words are extracted in word.

Step 714, multiple emotion words are calculated according to the weighted score of each emotion word in default dictionary to correspond to Emotion word score.

Step 716, the default word adjacent with emotion word is searched according to each emotion word.

Step 718, each sentence is calculated with emotion word score according to the weighted score of default word in default dictionary The corresponding emotion score of son.

Step 720, the affection index result of content of text to be feedback is calculated according to the emotion score of each sentence.

Specifically, after the emotion word in extracting each sentence, according to assigning each emotion word in default dictionary The corresponding emotion word score of multiple emotion words is calculated in weighted score.Further, on the basis of each emotion word The default words such as degree adverb, the negative word adjacent with emotion word are searched successively.It is default when finding degree adverb, negative word etc. After word, score value corresponding with default word is calculated to the weighted score for presetting word according to being assigned in default dictionary. After score value corresponding with default word is calculated, each sentence is calculated in the score of the score value of default word and emotion word The corresponding emotion score of son, the content of text to be feedback got is calculated further according to the corresponding emotion score of each sentence Affection index result.

Step 722, emotional semantic classification result and affection index result are compared according to the first weight distribution ratio, the second weight distribution It is weighted, obtains the sentiment analysis result of content of text to be feedback.

Specifically, by content of text perceptron treat feedback content of text be calculated the first weight distribution ratio and Second weight distribution ratio, the emotional semantic classification result according to the first weight distribution than comparing neural network output with the second weight distribution Corresponding fractional value and the affection index result obtained based on dictionary Algorithm Analysis are weighted, and are obtained in text to be feedback Hold corresponding sentiment analysis result.

Step 724, it treats feedback content of text and carries out syntactic type analysis, obtain the corresponding syntax of content of text to be feedback Type.

Step 726, when syntactic type in default cluster viewpoint library when the corresponding viewpoint syntactic type of viewpoint, according to word to The corresponding matching syntactic type of syntactic type is calculated in amount model.

Step 728, according to matching syntactic type match from viewpoint syntactic type obtain with match syntactic type it is corresponding Emotion viewpoint result.

As shown in figure 8, after getting content of text to be feedback, since content of text to be feedback is made of multiple words Sentence or article, in a practical situation, due to there are many combinations between word, then what multiple word groups were shaped as Sentence also has multiple.Therefore, syntactic type analysis need to be carried out to the content of text to be feedback got, obtains text to be feedback The corresponding syntactic type of content.Further, after the syntactic type for obtaining content of text to be feedback, when detecting text to be feedback When the corresponding viewpoint syntactic type of viewpoint of the syntactic type of this content in default viewpoint library, waited for according to the calculating of term vector model Feed back the similarity of the syntactic type and viewpoint syntactic type of content of text.When similarity is more than default similarity threshold, then It can be using the viewpoint syntactic type as the matching syntactic type of the syntactic type of content of text to be feedback.Therefore it is waited for instead getting After the matching syntactic type for presenting content of text, according to the corresponding sight of viewpoint from the viewpoint in cluster viewpoint library of matching syntactic type It is matched in point syntactic type.When successful match, the corresponding viewpoint of syntactic type can will be matched as in text to be feedback The emotion viewpoint result of appearance.

Step 730, it is replied in correlation database from default comment according to sentiment analysis result and emotion viewpoint result and chooses and wait for The corresponding target emotion feedback comments of content of text are fed back to reply.

As shown in figure 8, after getting sentiment analysis result and emotion viewpoint result, sentiment analysis result and emotion are seen Point result is as structural data, that is to say, that structural data is made of sentiment analysis result and emotion viewpoint result.According to Structural data is replied from default comment chooses comment reply sample corresponding with structural data in correlation database, comment is replied Reply sample in sample is replied as the corresponding target emotion feedback comments of content of text to be feedback.As shown in figure 9, Fig. 9 shows Go out the feedback result schematic diagram of the emotion feedback method perceived based on text in one embodiment.Such as：Content of text to be feedback is： " oh！Rapidly, this screen is very good-looking for this logistics.Mainly its price is low！", content of text to be feedback is as trained nerve The input of network obtains emotional semantic classification result：" like, praise ", acquisition preset emotional semantic classification result and are：" likes, praises Raise " fractional value be：" 10 points ".Content of text to be feedback obtains affection index result based on dictionary algorithm：" 5 points ", then The sentiment analysis result being weighted to emotional semantic classification result and affection index result by text perceptron is：" happiness Love is praised, 5 points ", wherein liking praising corresponding point based on what neural network obtained based on 5 points of the weight ratio that dictionary obtains The weight ratio of numerical value is more higher.Further, the emotion viewpoint result for obtaining content of text to be feedback is：" logistics is rapid ", root It is chosen from default comment reply correlation database according to sentiment analysis result and emotion viewpoint result corresponding with content of text to be feedback Target emotion feedback comments are replied：" so high evaluation do other too shy " returns target emotion feedback comments It is multiple to present to terminal user.

In the present embodiment, the output result based on dictionary and neural network is weighted by content of text perceptron, Obtain sentiment analysis result；Using sentiment analysis result and the emotion viewpoint result that gets as structural data from default comment It replys and gets the reply of target text content feed in library.The content of text for supporting analysis input in real time, is obtained with most fast speed Take the emotion feedback content with the content of text of input.Not only increase the matching of input content of text and emotion feedback content Rate, and improve the efficiency for obtaining emotion feedback content.

It should be understood that although each step in the flow chart of Fig. 1-7 is shown successively according to the instruction of arrow, These steps are not that the inevitable sequence indicated according to arrow executes successively.Unless expressly stating otherwise herein, these steps Execution there is no stringent sequences to limit, these steps can execute in other order.Moreover, at least one in Fig. 1-7 Part steps may include that either these sub-steps of multiple stages or stage are not necessarily in synchronization to multiple sub-steps Completion is executed, but can be executed at different times, the execution sequence in these sub-steps or stage is also not necessarily successively It carries out, but can either the sub-step of other steps or at least part in stage be in turn or alternately with other steps It executes.

In one embodiment, as shown in Figure 10, a kind of emotion feedback device 1000 perceived based on text is provided, is wrapped It includes：Content of text acquisition module 1002 to be feedback, weight distribution are than computing module 1004, emotional semantic classification result acquisition module 1006, affection index result computing module 1008, sentiment analysis interpretation of result module 1010, emotion viewpoint result acquisition module 1012 and feedback module 1014, wherein：

Content of text acquisition module 1002 to be feedback, for obtaining content of text to be feedback.

Weight distribution is treated feedback content of text than computing module 1004, for passing through content of text perceptron and is calculated First weight distribution ratio and the second weight distribution ratio.

Emotional semantic classification result acquisition module 1006 is obtained for content of text to be feedback to be inputted the neural network trained To emotional semantic classification result.

Affection index result computing module 1008, for multiple emotion word radix evidence in content of text to be feedback is default Corresponding affection index result is calculated in preset rules in dictionary.

Sentiment analysis interpretation of result module 1010, for comparing emotion according to the first weight distribution ratio, the second weight distribution Classification results and affection index result are weighted, and obtain the sentiment analysis result of content of text to be feedback.

Emotion viewpoint result acquisition module 1012, for obtaining the corresponding emotion viewpoint result of content of text to be feedback.

Feedback module 1014, for being replied in correlation database from default comment according to sentiment analysis result and emotion viewpoint result Target emotion feedback comments corresponding with content of text to be feedback are chosen to reply.

In one embodiment, as shown in figure 11, emotion viewpoint result acquisition module 1012 includes：Syntactic type analysis is single Member 1102, matching syntax type acquiring unit 1104 and emotion viewpoint result matching unit 1106, wherein

Syntactic type analytic unit 1102 carries out syntactic type analysis for treating feedback content of text, obtains to be feedback The corresponding syntactic type of content of text.

Syntax type acquiring unit 1104 is matched, for when syntactic type corresponding sight of viewpoint in default cluster viewpoint library When point syntactic type, the corresponding matching syntactic type of syntactic type is calculated according to term vector model.

Emotion viewpoint result matching unit 1106, for according to matching syntactic type viewpoint pair from default cluster viewpoint library The viewpoint syntactic type answered matches to obtain the corresponding emotion viewpoint result of content of text to be feedback.

In one embodiment, emotional semantic classification result acquisition module 1006 is additionally operable to treat feedback content of text and is divided It cuts, obtains multiple sentences；Each sentence is weighted according to special word, obtains the sentence after multiple weightings；To each sentence Son is split, and obtains multiple emotion words；It is calculated according to the weighted score of each emotion word in default dictionary multiple The corresponding emotion word score of emotion word；The default word adjacent with emotion word is searched according to each emotion word；According to It presets the weighted score of word in default dictionary and the corresponding emotion score of each sentence is calculated in emotion word score；According to The affection index result of content of text to be feedback is calculated in the emotion score of each sentence.

In one embodiment, the generation step in default cluster viewpoint library includes：Multiple comment samples are split, are obtained To the corresponding multiple words of each comment sample；Multiple words corresponding to each comment sample carry out part-of-speech tagging, obtain every The corresponding syntactic type of a comment sample；The corresponding syntax of each comment sample is calculated according to trained term vector model The similarity of type；When similarity is more than default similarity, the syntactic type that similarity is more than to default similarity is corresponding Multiple comment samples merge, and obtain comment sample set；The centre word for obtaining comment sample set, centre word is determined as Viewpoint is added in default cluster viewpoint library the viewpoint for commenting on sample set, generates default cluster viewpoint library, and wherein centre word is used In the type of definition comment sample set.

In one embodiment, the generation step of neural network includes：Obtain the comment sample marked；Sample will be commented on Trained term vector analysis model word2Vec is inputted, the corresponding term vector of comment sample is obtained；Using term vector as feature It inputs in neural network, neural network is trained, trained neural network is obtained.

In one embodiment, the generation step of comment reply correlation database includes：It obtains the comment marked and replys sample, It includes comment sample and corresponding reply sample that wherein sample is replied in comment；Sample is replied to comment by content of text perceptron First sample weight distribution ratio and the second sample weights distribution ratio is calculated；Sample input trained god is replied into comment Through network, sample emotional semantic classification result is obtained；It is default in dictionary according to presetting that multiple emotion word radix in sample are replied into comment Sample affection index result is calculated in rule；According to first sample weight distribution ratio, the second sample weights distribution ratio to sample Emotional semantic classification result and sample affection index result are weighted, and obtain the sample sentiment analysis result that sample is replied in comment；It obtains Comment is taken to reply the corresponding sample emotion viewpoint result of sample；By sample sentiment analysis result and emotion viewpoint result and will comment on It replys sample and is associated the reply sample of the comment after being associated with, sample generation comment is replied according to the comment after association and is replied Correlation database.

Specific restriction about the emotion feedback device perceived based on text may refer to above for based on text sense The restriction for the emotion feedback method known, details are not described herein.Each mould in the above-mentioned emotion feedback device based on text perception Block can be realized fully or partially through software, hardware and combinations thereof.Above-mentioned each module can be embedded in the form of hardware or independence In processor in computer equipment, can also in a software form it be stored in the memory in computer equipment, in order to Processor, which calls, executes the corresponding operation of the above modules.

In one embodiment, a kind of computer equipment is provided, which can be terminal, internal structure Figure is shown in Fig.12.The computer equipment includes the processor connected by system bus, memory, network interface, shows Display screen and input unit.Wherein, the processor of the computer equipment is for providing calculating and control ability.The computer equipment Memory includes non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system and computer Program.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The meter The network interface for calculating machine equipment is used to communicate by network connection with external terminal.When the computer program is executed by processor To realize a kind of emotion feedback method perceived based on text.The display screen of the computer equipment can be liquid crystal display or The input unit of electric ink display screen, the computer equipment can be the touch layer covered on display screen, can also be to calculate Button, trace ball or the Trackpad being arranged on machine equipment shell can also be external keyboard, Trackpad or mouse etc..

It will be understood by those skilled in the art that structure shown in Figure 12, only with the relevant part of application scheme The block diagram of structure, does not constitute the restriction for the computer equipment being applied thereon to application scheme, and specific computer is set Standby may include either combining certain components than more or fewer components as shown in the figure or being arranged with different components.

In one embodiment, a kind of computer equipment is provided, including memory, processor and storage are on a memory And the computer program that can be run on a processor, processor realize following steps when executing computer program：It obtains to be feedback Content of text；Feedback content of text is treated by content of text perceptron, and the first weight distribution ratio and the second weight point is calculated Proportioning；Content of text to be feedback is inputted to the neural network trained, obtains emotional semantic classification result；It will be in content of text to be feedback Multiple emotion word radix according to preset dictionary in preset rules corresponding affection index result is calculated；According to the first weight point Proportioning, the second weight distribution compare emotional semantic classification result and affection index result is weighted, and obtains content of text to be feedback Sentiment analysis result；Obtain the corresponding emotion viewpoint result of content of text to be feedback；According to sentiment analysis result and emotion viewpoint As a result it is replied from default comment and chooses target emotion feedback comments reply corresponding with content of text to be feedback in correlation database.

In one embodiment, a kind of computer readable storage medium is additionally provided, computer program is stored thereon with, it should Following steps are realized when program is executed by processor：Obtain content of text to be feedback；By content of text perceptron to be feedback The first weight distribution ratio and the second weight distribution ratio is calculated in content of text；Content of text to be feedback is inputted to the god trained Through network, emotional semantic classification result is obtained；By multiple emotion word radix in content of text to be feedback according to default rule in default dictionary Corresponding affection index result is then calculated；Emotional semantic classification result is compared according to the first weight distribution ratio, the second weight distribution It is weighted with affection index result, obtains the sentiment analysis result of content of text to be feedback；Obtain content of text pair to be feedback The emotion viewpoint result answered；It is replied in correlation database from default comment according to sentiment analysis result and emotion viewpoint result and chooses and wait for The corresponding target emotion feedback comments of content of text are fed back to reply.

In one embodiment, by multiple emotion word radix in content of text to be feedback according to preset rules in default dictionary The step of corresponding affection index result is calculated, including：It treats feedback content of text to be split, obtains multiple sentences； Each sentence is weighted according to special word, obtains the sentence after multiple weightings；Each sentence is split, is obtained more A emotion word；The corresponding emotion of multiple emotion words is calculated according to the weighted score of each emotion word in default dictionary Word score；The default word adjacent with emotion word is searched according to each emotion word；According to default word in default dictionary Weighted score and emotion word score the corresponding emotion score of each sentence is calculated；According to the emotion score of each sentence The affection index result of content of text to be feedback is calculated.

In one embodiment, the step of obtaining the emotion viewpoint result of content of text to be feedback answered, including：It treats anti- It presents content of text and carries out syntactic type analysis, obtain the corresponding syntactic type of content of text to be feedback；When syntactic type is default When clustering the corresponding viewpoint syntactic type of viewpoint in viewpoint library, the corresponding matching of syntactic type is calculated according to term vector model Syntactic type；It matches to obtain according to matching syntactic type corresponding viewpoint syntactic type of viewpoint from default cluster viewpoint library and wait for instead Present the corresponding emotion viewpoint result of content of text.

In one embodiment, the generation step in default cluster viewpoint library includes：Multiple comment samples are split, are obtained To the corresponding multiple words of each comment sample；Multiple words corresponding to each comment sample carry out part-of-speech tagging, obtain every The corresponding syntactic type of a comment sample；The corresponding syntax of each comment sample is calculated according to trained term vector model The similarity of type；When similarity is more than default similarity, the syntactic type that similarity is more than to default similarity is corresponding Multiple comment samples merge, and obtain comment sample set；The centre word for obtaining comment sample set, centre word is determined as Viewpoint is added in default cluster viewpoint library the viewpoint for commenting on sample set, generates default cluster viewpoint library, and wherein centre word is used In the type of definition comment sample.

In one embodiment, the generation step of comment reply correlation database includes：It obtains the comment marked and replys sample, It includes comment sample and corresponding reply sample that wherein sample is replied in comment；Sample is replied to comment by content of text perceptron First sample weight distribution ratio and the second sample weights distribution ratio is calculated；Sample input trained god is replied into comment Through network, sample emotional semantic classification result is obtained；It is default in dictionary according to presetting that multiple emotion word radix in sample are replied into comment Sample affection index result is calculated in rule；Sample emotion is compared according to first sample weight distribution ratio, the second weight distribution Classification results and affection index result carry out, and obtain the sample sentiment analysis result that sample is replied in comment；It obtains comment and replys sample This corresponding sample emotion viewpoint result；By sample sentiment analysis result and emotion viewpoint result and reply sample progress will be commented on Sample is replied in comment after being associated with, and replying sample according to the comment after association generates comment reply correlation database.

One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, being can be with Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer In read/write memory medium, the computer program is when being executed, it may include such as the flow of the embodiment of above-mentioned each method.Wherein, Any reference to memory, storage, database or other media used in each embodiment provided herein, Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM (PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms, Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..

Each technical characteristic of above example can be combined arbitrarily, to keep description succinct, not to above-described embodiment In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance Shield is all considered to be the range of this specification record.

The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art It says, under the premise of not departing from the application design, various modifications and improvements can be made, these belong to the protection of the application Range.Therefore, the protection domain of the application patent should be determined by the appended claims.

Claims

1. a kind of emotion feedback method based on text perception, the method includes：

Obtain content of text to be feedback；

The first weight distribution ratio and the second weight point are calculated to the content of text to be feedback by content of text perceptron Proportioning；

The content of text to be feedback is inputted into the neural network trained, obtains emotional semantic classification result；

Corresponding affection index result is calculated according to preset rules in dictionary are preset in the multiple emotion word radix；

The emotional semantic classification result and the affection index result are compared according to the first weight distribution ratio, the second weight distribution It is weighted, obtains the sentiment analysis result of the content of text to be feedback；

It replys to choose in correlation database from default comment according to the sentiment analysis result and the emotion viewpoint result and be waited for described The corresponding target emotion feedback comments of content of text are fed back to reply.

2. according to the method described in claim 1, it is characterized in that, described be split the content of text to be feedback, obtain To multiple emotion words in the content of text to be feedback, by the multiple emotion word radix according to preset rules in default dictionary Corresponding affection index is calculated as a result, including：

The content of text to be feedback is split, multiple sentences are obtained；

Each sentence is weighted according to special word, obtains the sentence after multiple weightings；

Each sentence is split, multiple emotion words are obtained；

Multiple emotion words are calculated according to the weighted score of each emotion word in the default dictionary to correspond to Emotion word score；

The default word adjacent with the emotion word is searched according to each emotion word；

Each institute is calculated with the emotion word score according to the weighted score of default word described in the default dictionary State the corresponding emotion score of sentence；

The affection index result of the content of text to be feedback is calculated according to the emotion score of each sentence.

3. according to the method described in claim 1, it is characterized in that, the emotion viewpoint answered for obtaining content of text to be feedback As a result, including：

Syntactic type analysis is carried out to the content of text to be feedback, obtains the corresponding syntax class of the content of text to be feedback Type；

When the syntactic type is in the corresponding viewpoint syntactic type of viewpoint in presetting cluster viewpoint library, according to term vector model meter Calculation obtains the corresponding matching syntactic type of the syntactic type；

Matched from the viewpoint syntactic type according to the matching syntactic type obtain it is corresponding with the matching syntactic type Emotion viewpoint result.

4. according to the method described in claim 3, it is characterized in that, the generation step in the default cluster viewpoint library includes：

Multiple comment samples are split, the corresponding multiple words of each comment sample are obtained；

The multiple word corresponding to each comment sample carries out part-of-speech tagging, obtains each comment sample correspondence Syntactic type；

The similarity of the corresponding syntactic type of each comment sample is calculated according to trained term vector model；

When the similarity is more than default similarity, the syntactic type that the similarity is more than to default similarity is corresponding more A comment sample merges, and obtains comment sample set；

The centre word is determined as the viewpoint of the comment sample set by the centre word for obtaining the comment sample set, will The viewpoint is added in the default cluster viewpoint library, generates the default cluster viewpoint library, wherein the centre word is for fixed The type of the justice comment sample set.

5. according to the method described in claim 1, it is characterized in that, the generation step of the neural network includes：

Obtain the comment sample marked；

The comment sample is inputted into trained term vector analysis model word2Vec, obtains the corresponding word of the comment sample Vector；

It is inputted the term vector as feature in the neural network, the neural network is trained, is trained Neural network.

6. according to the method described in claim 1, it is characterized in that, the generation step that correlation database is replied in the comment includes：

It obtains the comment marked and replys sample, wherein it includes comment sample and corresponding reply sample that sample is replied in the comment This；

Sample is replied to the comment by the content of text perceptron, first sample weight distribution ratio and second is calculated Sample weights distribution ratio；

Trained neural network described in sample input is replied into the comment, obtains sample emotional semantic classification result；

Multiple emotion word radix in sample are replied into the comment, sample is calculated according to preset rules in the default dictionary Affection index result；

According to the first sample weight distribution ratio, the second sample weights distribution ratio to the sample emotional semantic classification result and described Sample affection index result is weighted, and obtains the sample sentiment analysis result that sample is replied in the comment；

It obtains the comment and replys the corresponding sample emotion viewpoint result of sample；

The sample sentiment analysis result and the emotion viewpoint result are associated to obtain with by comment reply sample Sample is replied in comment after association, and replying sample according to the comment after association generates the comment reply correlation database.

7. a kind of emotion feedback device based on text perception, described device include：

Weight distribution is calculated first than computing module, for passing through content of text perceptron to the content of text to be feedback Weight distribution ratio and the second weight distribution ratio；

Emotional semantic classification result acquisition module obtains feelings for the content of text to be feedback to be inputted the neural network trained Feel classification results；

Affection index result computing module obtains the text to be feedback for being split to the content of text to be feedback Corresponding feelings are calculated according to preset rules in dictionary are preset in the multiple emotion word radix by multiple emotion words in content Feel index results；

Sentiment analysis interpretation of result module, for comparing the emotion according to the first weight distribution ratio, the second weight distribution Classification results and the affection index result are weighted, and obtain the sentiment analysis result of the content of text to be feedback；

Feedback module, for being replied in correlation database from default comment according to the sentiment analysis result and the emotion viewpoint result Choose target emotion feedback comments reply corresponding with the content of text to be feedback.

8. device according to claim 7, which is characterized in that the emotion viewpoint result acquisition module includes：

Syntactic type analytic unit obtains described to be feedback for carrying out syntactic type analysis to the content of text to be feedback The corresponding syntactic type of content of text；

Syntax type acquiring unit is matched, for when the syntactic type is in the default corresponding viewpoint sentence of viewpoint in clustering viewpoint library When method type, the corresponding matching syntactic type of the syntactic type is calculated according to term vector model；

Emotion viewpoint result matching unit, for according to the matching syntactic type from the default cluster viewpoint library viewpoint pair The viewpoint syntactic type answered matches to obtain the corresponding emotion viewpoint result of the content of text to be feedback.

9. a kind of computer equipment, including memory, processor and storage are on a memory and the meter that can run on a processor Calculation machine program, which is characterized in that the processor realizes any one of claim 1 to 6 institute when executing the computer program The step of stating method.

10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program The step of method according to any one of claims 1 to 6 is realized when being executed by processor.