CN112527866A - Stock trend prediction method and system based on text abstract emotion mining - Google Patents
Stock trend prediction method and system based on text abstract emotion mining Download PDFInfo
- Publication number
- CN112527866A CN112527866A CN202011497913.0A CN202011497913A CN112527866A CN 112527866 A CN112527866 A CN 112527866A CN 202011497913 A CN202011497913 A CN 202011497913A CN 112527866 A CN112527866 A CN 112527866A
- Authority
- CN
- China
- Prior art keywords
- stock
- emotion
- news
- trend
- text
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000008451 emotion Effects 0.000 title claims abstract description 121
- 238000000034 method Methods 0.000 title claims abstract description 36
- 238000005065 mining Methods 0.000 title claims abstract description 20
- 230000008859 change Effects 0.000 claims abstract description 29
- 238000004364 calculation method Methods 0.000 claims abstract description 25
- 239000013598 vector Substances 0.000 claims abstract description 16
- 238000012549 training Methods 0.000 claims description 13
- 230000007246 mechanism Effects 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 11
- 230000002996 emotional effect Effects 0.000 claims description 6
- 230000010415 tropism Effects 0.000 claims description 5
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000011160 research Methods 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000015654 memory Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013481 data capture Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007787 long-term memory Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000007935 neutral effect Effects 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000012163 sequencing technique Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
- G06F16/2465—Query processing support for facilitating data mining operations in structured databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Accounting & Taxation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Finance (AREA)
- Development Economics (AREA)
- Economics (AREA)
- Marketing (AREA)
- Strategic Management (AREA)
- Technology Law (AREA)
- General Business, Economics & Management (AREA)
- Fuzzy Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a stock trend prediction method and a system based on text abstract emotion mining, which comprises the following steps: s1, obtaining a plurality of news data related to the stock; s2, obtaining the text abstract of each news through news data; s3, extracting emotion words in each text abstract according to a pre-established emotion word library, expressing the intensity of emotion according to the emotion words, and scoring the emotion of each text abstract; s4, inputting the emotion scores of the text abstracts as feature vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock is in an upward trend; if the calculation result is less than zero, the stock is in a falling trend. By extracting the news text abstract and carrying out emotion mining on the text abstract, the information which influences the fluctuation trend of the stock market is effectively obtained, so that the stock fluctuation prediction is not limited to the information of the current stock, and the stock trend can be more accurately predicted from more aspects.
Description
Technical Field
The invention relates to a stock trend prediction method and system based on text abstract emotion mining, and belongs to the technical field of artificial intelligence.
Background
The Chinese stock market is an emerging security market, the market structure and mechanism are not mature, and the process of continuous exploration and improvement is performed. Compared with the western mature market, the investment of the stock market in China is mainly performed by small and medium-sized investors, the information acquisition and analysis capability and the cognitive capability of the investors are not uniformly distributed, and the information asymmetry degree among the investors is higher than that of the western mature capital market. For the new market in China, the investors have relatively poor experience and more irrational behaviors, and the emotion of the investors can be enlarged and the fluctuation of stock market is aggravated by excessive participation of mainstream media. Research has shown that emotional tendency of internet financial news influences decision-making and behaviors of investors, thereby causing market fluctuation.
In recent years, many researchers have participated in stock forecasting with news sentiment as an input feature. When reading news information, the stock investor can artificially extract main information and avoid the interference of irrelevant information. However, there is often a bias when automatically extracting news sentiment tendencies using text sentiment analysis techniques. Especially, when the emotion polarity changes in the news text, the primary and secondary relations cannot be distinguished by the model, and all emotions are calculated and output together. At present, a plurality of scholars both domestic and abroad propose a solution for automatic text summarization. By extracting the news in the abstract mode and taking out effective information contained in the news, the problems of information overload and news emotion polarity change can be effectively solved, and the emotion tendency of the information can be extracted more accurately. Therefore, the application of the existing natural language processing technology and stock forecasting technology to deeply research the influence of news information abstract on stock fluctuation and forecast the stock fluctuation trend has theoretical and practical significance on investment decision of investors, improvement of market construction and the like.
Disclosure of Invention
Aiming at the defects of the prior art, the invention aims to provide a method and a system for predicting the stock trend based on text abstract emotion mining, which effectively obtain information which can affect the stock market fluctuation trend by extracting a news text abstract and performing emotion mining on the text abstract, so that the stock fluctuation prediction is not only limited to the current stock information, but also can more accurately predict the stock trend from more aspects.
In order to achieve the purpose, the invention provides a stock trend prediction method based on text abstract emotion mining, which comprises the following steps of: s1, obtaining a plurality of news data related to the stock; s2, obtaining the text abstract of each news through news data; s3, extracting emotion words in each text abstract according to a pre-established emotion word library, and scoring the emotion of each text abstract according to the strength of emotion word expression emotion; s4, inputting the emotion scores of the text abstracts as characteristic vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock shows an upward trend; if the calculation result is less than zero, the stock is in a falling trend.
Further, the step S1 of obtaining several news data related to the stock includes: selecting a stock to be researched, acquiring at least one of stock code abbreviation, date, opening price, highest price, lowest price, closing price, fluctuation width, volume of trades, amount of trades, hand-off rate, A-stock market value, total market value, market profit rate and market selling rate in a certain period of time, generating a retrieval keyword, selecting a reliable news source, and retrieving news data related to the stock from the news source.
Further, the specific operation procedure of step S2 is: s2.1, performing word segmentation on the extracted news data, removing punctuation marks, replacing unrelated entities with labels, and converting words in the news data into dictionary indexes; s2.2, training a news source selection model by using the extracted news data to obtain a reliable news source; s2.3, news data are extracted from reliable news sources again, and text summaries of various news are obtained through a Seq2Seq model of an Attention mechanism.
Further, the training process of the news source selection model is as follows: retrieving news data related to the stock to be predicted in a certain historical time from a certain news source, acquiring the actual change trend of the stock to be predicted in the historical time, predicting the change trend of the stock to be predicted by using the news data, comparing the predicted change trend with the actual change trend, and scoring according to the comparison result; and performing the above operation on all the news sources to be selected to obtain the scores of all the news sources to be selected, and selecting the news sources with the scores exceeding the threshold value as reliable news sources.
Further, the process of obtaining the text abstract of each news through the Seq2Seq model of the Attention mechanism is as follows: the news data is used as an input vector sequence and input into a Seq2Seq model of an Attention mechanism to obtain the generation probability of each word, so that the generation probability of all the words of the news data is obtained, and the words with high generation probability are selected to generate the text abstract of the news.
Further, the emotion vocabulary library in step S3 includes: emotional words, the term of art with tropism in financial news, mean negative words and degree adverbs.
Further, the specific operation procedure of step S3 is: the method comprises the steps of dividing emotion words in an emotion word bank and professional terms with tendentiousness in financial news into two types of positive emotion expression and negative emotion expression, setting scores of the emotion words in each emotion word bank and the professional terms with tendentiousness in the financial news, adding corresponding scores if positive emotions are expressed, subtracting corresponding scores if negative emotions are expressed, dividing text abstracts of each news data into a plurality of sentence clusters, and obtaining scores of the sentence clusters by combining negative words and degree adverbs.
Further, the rule of scoring each sentence cluster obtained by combining the negative word and the degree adverb is as follows: the score of a negative word appearing only is Un ═ 0.1 XCn-1)CnXs, score for only the occurrence of degree adverb: u shapes＝∏LsX S, when a negative word and a degree adverb exist at the same time, if the negative word is in front, adding the total score of the negative word and the total score of the degree adverb; if the negative word is later, multiplying the total fraction of the negative word and the total fraction of the degree adverb, Un is the total fraction of the negative word, Us is the total fraction of the degree adverb, Cn is the score of the negative word, Ls is the score of the degree adverb, and S is the score of a professional term with tropism in the news without the emotion word and the financial news.
Further, the specific operation steps of step S4 are: 4.1, selecting a time window, and collecting stock ticket historical change trend data in the time window; 4.2, the emotion scores of the text abstracts are used as feature vectors and stock historical change trend data to be input into an LSTM model, and the LSTM model is trained to obtain a stock training model containing emotion features; 4.3 inputting the emotion scores of the text abstracts for judging the stock trend in the stock training model containing the emotion characteristics as characteristic vectors and stock historical change trend data, and calculating the variation of stock belts in a prediction time period; 4.4 judging whether the variation is larger than zero, if the variation is larger than or equal to zero, the stock is in an upward trend; if the variation is less than zero, the stock is in a falling trend.
The invention also discloses a stock trend prediction system based on text abstract emotion mining, which comprises the following steps: the acquisition module is used for acquiring a plurality of news data related to the stock; the abstract generating module is used for obtaining text abstract of each news through news data; the emotion scoring module is used for extracting emotion words in each text abstract according to a pre-established emotion word bank, expressing the intensity of emotion according to the emotion words and scoring the emotion of each text abstract; the trend judgment module is used for inputting the emotion scores of the text abstracts as characteristic vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock shows an upward trend; if the calculation result is less than zero, the stock is in a falling trend.
Due to the adoption of the technical scheme, the invention has the following advantages:
1. the method comprises the steps of collecting and sorting news texts based on stocks, constructing a large text abstract database, and calculating the sentiment value of the abstract; the automatic text summarization and emotion value calculation are applied to the deep learning prediction model, so that various input information of the model is perfected, and the representation capability of the model is improved.
2. The invention combines the emotion value of the text abstract and predicts the stock fluctuation through the LSTM network, thereby improving the accuracy of the prediction effect. Therefore, the method can be widely applied to the field of stock forecasting application.
Drawings
FIG. 1 is a flow chart of a method for predicting a stock fluctuation trend based on automatic text summarization and emotion mining in an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a Seq2Seq model based on an attribution mechanism in an embodiment of the present invention;
FIG. 3 is a schematic diagram of the structure of the LSTM prediction model in an embodiment of the present invention.
Detailed Description
The present invention is described in detail by way of specific embodiments in order to better understand the technical direction of the present invention for those skilled in the art. It should be understood, however, that the detailed description is provided for better understanding of the present invention only and should not be taken as limiting the present invention. In describing the present invention, it is to be understood that the terminology used is for the purpose of description only and is not intended to be indicative or implied of relative importance.
Example one
The embodiment discloses a stock trend prediction method based on text abstract emotion mining, which comprises the following steps of:
s1 obtains several news data related to the stock.
Obtaining several stock related news data includes: determining a stock object according to the volume of the stock, the time interval of stock listing and the attention degree, and acquiring the market data of the selected stock; and selecting a research object. Firstly, whether the stock has valid data, related news amount, stock volume ranking and stock income situation ranking in the research time range needs to be considered to ensure that the researched stock has higher attention, so that the investment psychology and investment decision of investors can be influenced when the news sentiment value changes. After the selected stocks are determined, the selected stock information is exported through a database, stocks to be researched are selected, at least one of stock code abbreviation, date, opening price, highest price, lowest price, closing price, fluctuation amplitude, volume of trades, amount of trades, hand-changing rate, A-stock market value, total market value, market profitability and market selling rate in a certain period of time of the stocks is obtained, retrieval keywords are generated, reliable news sources are selected, and news data related to the stocks are retrieved from the news sources. The reliable news source can be determined by comparing all the large financial websites according to the influence, the importance and the acquirability, and the information source for capturing news is determined. The content of each large financial portal website is analyzed and compared mainly according to the influence, the importance and the acquirability, a news acquisition range is defined, and a representative stock news portal website is selected for data capture. In this embodiment, a certain amount of news information is selected from alternative relatively common and authoritative news sources for training, so that a news source selection model is established, and more representative and more authentic news information can be extracted. The news source can be updated at any time according to the prediction accuracy, and can also be updated regularly. After a news source is selected, a crawler script is written according to a stock object to be researched by adopting a character string matching mode, and news information in a period of time is captured by using a script crawler frame through the code, name and abbreviation of the stock as key words. The captured content includes a news title, a release time, release content, a website, and the like, thereby composing news data.
S2 obtains a text summary of each news item from the news data.
The specific operation process is as follows:
s2.1, performing word segmentation on the extracted news data, removing punctuation marks, and uniformly replacing unrelated entities such as Date, Time, Name, Number and the like with labels such as Date, Time, Name, Number and the like. Constructing a dictionary with the length of 60000 based on news data, and converting words in the news data into dictionary indexes;
s2.2, training a news source selection model by using the extracted news data to obtain a reliable news source; the training process of the news source selection model is as follows: retrieving news data related to the stock to be predicted in a certain historical time from a certain news source, acquiring the actual change trend of the stock to be predicted in the historical time, predicting the change trend of the stock to be predicted by using the news data, comparing the predicted change trend with the actual change trend, and scoring according to the comparison result; and performing the above operation on all the news sources to be selected to obtain scores of all the news sources to be selected, and selecting the news sources with the scores exceeding the threshold value as reliable news sources.
S2.3, news data are extracted from reliable news sources again, and text summaries of various news are obtained through a Seq2Seq model of an Attention mechanism.
The process of obtaining the text abstract of each news through the Seq2Seq model of the Attention mechanism is as follows: and inputting news data serving as an input vector sequence into a Seq2Seq model of an Attention mechanism to obtain the generation probability of each word, thereby obtaining the generation probability of all words of the news data, and selecting the words with high generation probability, thereby generating a text abstract of news.
As shown in fig. 2, the Attention mechanism is a channel connecting an encoding layer (Encoder) and a decoding layer (Decoder). Since the hidden layer of each recurrent neural network RNN unit is preserved in the Encode, the hidden layer h at the t-th time step of the Encode is assumedtHidden layer htIs recorded as ht＝f(ht-1,yt-1C) including input sequence information c, output y of previous time instantt-1And a hidden layer of the previous momentht-1Let the hidden layer at the t-th time step be htThen, the weight of the ith word of the input text of the corresponding coding part at the t-th time step can be calculated, and softmax calculation is carried out to obtain the final weight alphati. The Decoder part hides the information of all layers (h) in the input sequence1,h2,…ht) Carrying out weighted summation, and selecting the most appropriate context information c with the currently required output word itAnd according to the output sequence information CtObtaining a hidden layer St of the Decoder at the t time step, and obtaining the output y of the t word according to the hidden layer Stt. Wherein, ytThe generation probability of (a) is: p (y)t|y(t-1),yt-2,…,y1,c)＝g(ht,yt-1,c)。
S3, extracting the emotion words in each text abstract according to the pre-established emotion word library, and scoring the emotion of each text abstract according to the strength of the emotion expressed by the emotion words.
The emotion word bank comprises: emotional words, the term of art with tropism in financial news, mean negative words and degree adverbs. Constructing a professional term with tendency in an emotion dictionary and financial news, and expressing negative words and degree adverbs comprises the following steps:
and establishing a basic emotion word bank by using HowNet emotion words and evaluation words, fusing a Chinese emotion word bank of the general organization without neutral words, a Chinese emotion dictionary of Taiwan university which is deleted and is not suitable for being directly used for emotion analysis words and partial news emotion words on the basis, and establishing a relatively complete emotion word bank. Secondly, by extracting comment information from each big forum and extracting sentiment words by sequencing news texts, a representative professional word library is collected, and 3 experts (security practitioners and stockholders) in the field are invited to score respectively.
And then, providing a negative word category, classifying the negative words, and selecting to obtain a negative word dictionary by combining the news word frequency statistical condition and the daily use condition.
And finally, classifying the degree adverbs, constructing a degree dictionary by combining degree level words and news word frequency statistical conditions in HowNet, and endowing different degree words with different scores for emotion calculation.
The specific operation process of step S3 is: dividing emotion words in an emotion word bank and professional terms with tendentiousness in financial news into two categories of positive emotion expression and negative emotion expression, setting scores of the emotion words in each emotion word bank and the professional terms with tendentiousness in the financial news, adding corresponding scores if positive emotion is expressed, subtracting corresponding scores if negative emotion is expressed, dividing text abstracts of each news data into a plurality of sentence clusters, and obtaining scores of the sentence clusters by combining negative words and degree adverbs. Given that news typically uses a more regular written language, while spoken inverted sentences are less common, this document uses simple logic to process ambiguous words. The polysemous words before the emotional words are regarded as degree words; the last polysemous word is regarded as the sentiment word, and other polysemous words are regarded as the degree word.
The score rule for obtaining each sentence cluster by combining the negative words and the degree adverbs is as follows: the score of a negative word appearing only is Un ═ 0.1 XCn-1)CnXs, score for only the occurrence of degree adverb: u shapes＝∏LsX S, when a negative word and a degree adverb exist at the same time, if the negative word is in front, adding the total score of the negative word and the total score of the degree adverb; if the negative word is later, multiplying the total score of the negative word and the total score of the degree adverb, Un is the total score of the negative word, Us is the total score of the degree adverb, Cn is the score of the negative word, Ls is the score of the degree adverb, and S is the score of a professional term which does not include the emotion word and the tendency in the financial news. Specific scoring criteria are shown in table 1.
TABLE 1 score calculation criteria for sentiment values
Averaging according to all the news emotion values of the day to obtain the news emotion value of the day. All the news emotion values can also be input into a stock forecasting model for training.
S4, inputting the emotion scores of the text abstracts as feature vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock is in an upward trend; if the calculation result is less than zero, the stock is in a falling trend.
The specific operation steps of step S4 are:
4.1 selecting a time window and collecting historical stock change trend data in the time window.
4.2 the sentiment score of each text abstract is used as a characteristic vector and stock historical change trend data to be input into a long-term and short-term memory network model, namely an LSTM model, and the LSTM model is trained to obtain a stock training model containing sentiment characteristics.
The LSTM model differs from the recurrent neural network model RNN mainly in that a processor is added to the algorithm to judge whether information is useful or not, and the structure acted by the processor is called a cell. Three doors, namely an input door, a forgetting door and an output door, are arranged in one cell. A message enters the LSTM network and may be determined to be useful based on rules. Only the information which accords with the algorithm authentication is left, and the information which does not accord with the algorithm authentication is forgotten through a forgetting door.
Wherein, the model of forgetting the door does: f. oft＝σ(Wf[xt,ht-1]+bf) Wherein f ist∈[0,1]The selection weight of the node pair at the time t to the cell at the time t-1 is shown, bfBiasing term for forgetting gate, WfIs a parameter, ht-1To representt-1Hidden layer state information of a node, a nonlinear function σ (x) is 1/(1+ e)-x) (ii) a Input Gate model it＝σ(Wi[xt,ht-1]+bi) Wherein it∈[0,1]The selection weight of the node at time t to the current node information, biFor input of offset terms of gates, WiAs a parameter, the nonlinear function σ (x) is 1/(1+ e)-x) (ii) a Current node input information
Wherein
Is the bias term.
Outputting memory information
Wherein C ist-1Representing the memory information of the t-1 node.
Output gate ot＝σ(Wo[xt,ht-1]+bo) Wherein o ist∈[0,1]Selection weight of node cell memory information representing time t, boIs the bias term of the output gate; woHidden layer state h at time t as parametert＝ot·tanh(Ct) (ii) a The hidden layer is connected with the output layer in a full mode, and two probability values, namely the probability with the variable quantity larger than zero or smaller than zero, are output after being mapped by a softmax nonlinear function
4.3 inputting the emotion scores of the text abstracts for judging the stock trend in the stock training model containing the emotion characteristics as characteristic vectors and stock historical change trend data, and calculating the variation of the stock belts in the prediction time period.
4.4, judging whether the variation is larger than zero, if the variation is larger than or equal to zero, the stock is in an upward trend; if the variation is less than zero, the stock is in a falling trend. In the process of predicting the stock fluctuation trend, the predicted value of the next day is based on stock history data, and the LSTM can directly process sequences with any length, so that the stock prediction requirement can be met. LSTM has the ability to remove or add information to the state of a cell through a well-designed structure called a "gate".
Example two
Based on the same inventive concept, the embodiment discloses a stock trend prediction system based on text abstract emotion mining, which comprises:
the acquisition module is used for acquiring a plurality of news data related to the stock;
the abstract generating module is used for obtaining text abstract of each news through news data; the emotion scoring module is used for extracting emotion words in the text abstracts according to a pre-established emotion word bank and scoring the emotion of each text abstract according to the strength of emotion words expressing emotion;
the trend judgment module is used for inputting the emotion scores of the text abstracts as characteristic vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock is in an increasing trend; if the calculation result is less than zero, the stock is in a falling trend.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (10)
1. A stock trend prediction method based on text abstract emotion mining is characterized by comprising the following steps:
s1, obtaining a plurality of news data related to the stock;
s2, obtaining the text abstract of each news through the news data;
s3, extracting emotion words in each text abstract according to a pre-established emotion word bank, and scoring the emotion of each text abstract according to the strength of emotion expressed by the emotion words;
s4, inputting the emotion scores of the text abstracts as feature vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock shows an upward trend; if the calculation result is less than zero, the stock is in a falling trend.
2. The method for predicting the trend of stocks based on emotion mining of text summary as claimed in claim 1, wherein the step S1 for obtaining several news data related to stocks includes: selecting a stock to be researched, acquiring at least one of stock code abbreviation, date, opening price, highest price, lowest price, closing price, fluctuation breadth, volume of trades, amount of trades, hand-off rate, A-stock circulation market value, total market value, market profit rate and market selling rate of the stock within a certain time, generating a retrieval keyword, selecting a reliable news source, and retrieving news data related to the stock from the news source.
3. The method for predicting the trend of stocks based on emotion mining of text summary as claimed in claim 1, wherein the specific operation procedure of said step S2 is as follows:
s2.1, performing word segmentation on the extracted news data, removing punctuation marks, replacing unrelated entities with labels, and converting words in the news data into dictionary indexes;
s2.2, training a news source selection model by using the extracted news data to obtain a reliable news source;
s2.3, news data are extracted from the reliable news source again, and text abstracts of news are obtained through a Seq2Seq model of the Attention mechanism.
4. The method of claim 3, wherein the news source selection model is trained as follows: retrieving news data related to the stock to be predicted in a certain historical time from a certain news source, acquiring the actual change trend of the stock to be predicted in the historical time, predicting the change trend of the stock to be predicted by using the news data, comparing the predicted change trend with the actual change trend, and scoring according to the comparison result; and performing the above operation on all the news sources to be selected to obtain the scores of all the news sources to be selected, and selecting the news sources with the scores exceeding the threshold value as reliable news sources.
5. The method for predicting the stock trend based on the emotion mining of text abstract as claimed in claim 3, wherein the process of obtaining the text abstract of each news through the Seq2Seq model of the Attention mechanism is: and inputting the news data serving as an input vector sequence into a Seq2Seq model of an Attention mechanism to obtain the generation probability of each word so as to obtain the generation probability of all the words of the news data, and selecting the words with high generation probability so as to generate a text abstract of the news.
6. The method for predicting a stock trend based on emotion mining of text abstract as claimed in any one of claims 1 to 5, wherein said emotion lexicon in step S3 includes: emotional words, the term of art with tropism in financial news, mean negative words and degree adverbs.
7. The method for predicting the trend of stocks based on emotion mining of text summary as claimed in claim 6, wherein the specific operation procedure of step S3 is as follows: dividing emotion words in the emotion word stock and professional terms with tendencies in financial news into two categories of positive emotion expression and negative emotion expression, setting scores of the emotion words in each emotion word stock and the professional terms with tendencies in the financial news, adding corresponding scores if positive emotions are expressed, subtracting corresponding scores if negative emotions are expressed, dividing text abstracts of each news data into a plurality of sentence clusters, and combining negative words and degree adverbs to obtain scores of the sentence clusters.
8. The method as claimed in claim 7, wherein the rule for obtaining the score of each sentence cluster by combining the negative words and the adverbs is: the score of a negative word appearing only is Un ═ 0.1 XCn-1)CnXs, score for only the occurrence of degree adverb: u shapes＝∏LsX S, when a negative word and a degree adverb exist at the same time, if the negative word is in front, adding the total score of the negative word and the total score of the degree adverb; if the negative word is later, multiplying the total fraction of the negative word and the total fraction of the degree adverb, and judging whether Un is definiteThe total score of words, Us is the total score of the degree adverb, Cn is the score of the negation word, Ls is the score of the degree adverb, and S is the score of the professional term with tropism in the news excluding the emotional word and the financial news.
9. The method for predicting the tendency of stocks based on emotion mining of text summary according to any of claims 1-5, wherein the specific operation steps of said step S4 are:
4.1 selecting a time window and collecting historical stock change trend data in the time window;
4.2 the sentiment score of each text abstract is used as a feature vector and stock historical change trend data to be input into an LSTM model, and the LSTM model is trained to obtain a stock training model containing sentiment features;
4.3 inputting the emotion scores of the text abstracts for judging the stock trend in the stock training model containing the emotion characteristics as characteristic vectors and stock historical change trend data, and calculating the variation of the stock belts in a prediction time period;
4.4, judging whether the variation is larger than zero, if so, the stock is in an upward trend; if the variable quantity is smaller than zero, the stock is in a falling trend.
10. A stock trend prediction system based on text abstract emotion mining, which is characterized by comprising:
the acquisition module is used for acquiring a plurality of news data related to the stock;
the abstract generating module is used for obtaining the text abstract of each news through the news data;
the emotion scoring module is used for extracting emotion words in the text abstracts according to a pre-established emotion word bank, and scoring the emotion of the text abstracts according to the strength of emotion expressed by the emotion words;
the trend judgment module is used for inputting the emotion scores of the text abstracts as feature vectors and historical stock change trend data into a pre-established stock prediction model for calculation, and if the calculation result is greater than or equal to zero, the stock shows an upward trend; if the calculation result is less than zero, the stock is in a falling trend.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010516651.1A CN111723127A (en) | 2020-06-09 | 2020-06-09 | Stock trend prediction method and system based on text abstract emotion mining |
| CN2020105166511 | 2020-06-09 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112527866A true CN112527866A (en) | 2021-03-19 |
Family
ID=72566200
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010516651.1A Pending CN111723127A (en) | 2020-06-09 | 2020-06-09 | Stock trend prediction method and system based on text abstract emotion mining |
| CN202011497913.0A Pending CN112527866A (en) | 2020-06-09 | 2020-12-17 | Stock trend prediction method and system based on text abstract emotion mining |
Family Applications Before (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010516651.1A Pending CN111723127A (en) | 2020-06-09 | 2020-06-09 | Stock trend prediction method and system based on text abstract emotion mining |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN111723127A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052706A (en) * | 2021-04-30 | 2021-06-29 | 中国银行股份有限公司 | Fund risk grade prediction method and device |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI811580B (en) * | 2020-11-04 | 2023-08-11 | 合作金庫商業銀行股份有限公司 | Financial information provisioning system and method for providing financial information |
-
2020
- 2020-06-09 CN CN202010516651.1A patent/CN111723127A/en active Pending
- 2020-12-17 CN CN202011497913.0A patent/CN112527866A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113052706A (en) * | 2021-04-30 | 2021-06-29 | 中国银行股份有限公司 | Fund risk grade prediction method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111723127A (en) | 2020-09-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110335B (en) | Named entity identification method based on stack model | |
| CN112001186A (en) | Emotion classification method using graph convolution neural network and Chinese syntax | |
| CN111914087B (en) | Public opinion analysis method | |
| CN103778215A (en) | Stock market forecasting method based on sentiment analysis and hidden Markov fusion model | |
| CN107403017A (en) | A kind of method that real-time news of intellectual analysis influences on financial market | |
| CN110909529B (en) | User emotion analysis and prejudgment system of company image promotion system | |
| CN111242679A (en) | Sales forecasting method based on product review viewpoint mining | |
| CN112256866A (en) | Text fine-grained emotion analysis method based on deep learning | |
| CN112527866A (en) | Stock trend prediction method and system based on text abstract emotion mining | |
| CN115658890A (en) | Chinese comment classification method based on topic-enhanced emotion-shared attention BERT model | |
| Zhang et al. | A hybrid neural network approach for fine-grained emotion classification and computing | |
| CN114692623A (en) | Emotion analysis method for environment network public sentiment | |
| Wang et al. | Joint Learning on Relevant User Attributes in Micro-blog. | |
| CN113505207B (en) | Machine reading understanding method and system for financial public opinion research report | |
| Jasim et al. | Analyzing Social Media Sentiment: Twitter as a Case Study | |
| CN114020901A (en) | Financial public opinion analysis method combining topic mining and emotion analysis | |
| CN112163951A (en) | Stock prediction method based on investor psychological mood | |
| Agbesi et al. | Multichannel 2D-CNN Attention-Based BiLSTM Method for Low-Resource Ewe Sentiment Analysis | |
| Singh et al. | Sentiment Analysis on IMDB Review Dataset | |
| Toktarova et al. | Offensive Language Identification in Low Resource Languages using Bidirectional Long-Short-Term Memory Network | |
| CN112836011B (en) | Analysis management system construction method and model construction based on big data application | |
| Wang et al. | Novel approaches to sentiment analysis for stock prediction | |
| CN114511345B (en) | Sales prediction method based on policy-public opinion-purchase two-stage deep learning | |
| Akshara et al. | An Effective Model to Prognosticate the Stock Market Tendency Using Social Media | |
| CN115759116A (en) | Aspect-level emotion analysis method based on double-graph convolutional network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210319 |