CN110555091A

CN110555091A - Associated word generation method and device based on word vectors

Info

Publication number: CN110555091A
Application number: CN201910805091.9A
Authority: CN
Inventors: 沈之锐
Original assignee: Shaoguan Qizhi Information Technology Co Ltd
Current assignee: Shaoguan Qizhi Information Technology Co Ltd
Priority date: 2019-08-29
Filing date: 2019-08-29
Publication date: 2019-12-10

Abstract

the invention provides a method and a device for generating associative words based on word vectors. The method comprises the following steps: acquiring a word to be generated by an associative word, decomposing the word into syllables according to the phonetic symbol of the word, taking the pronunciation of the syllables as a prefix, looking up a table to acquire the pinyin of the syllables and acquiring a Chinese character corresponding to the pinyin; aiming at the Chinese characters, combining words, verifying the combined words and obtaining candidate words; acquiring the part of speech of the candidate words, and preferentially selecting the candidate words according to the part of speech; according to the word vector technology, matching candidate words and similarity among the words are obtained, and the matching similarity is sequenced; and acquiring word Chinese explanation, word similarity between the word Chinese explanation and the candidate words, and reordering collocation similarity. Finally obtaining the associative word of the word; the method can make the computer replace the trouble of manually generating harmonic associated words, improve the generation efficiency of the associated words and help people to better memorize the words.

Description

Associated word generation method and device based on word vectors

Technical Field

the invention relates to the technical field of computer application, in particular to a method and a device for generating an associative word based on a word vector.

background

When people remember words, more methods for assisting people to remember the words are often needed. The method for associating the word pronunciation with the Chinese pronunciation and the semantics is currently applied to a plurality of memory methods. Even many training institutions use this method to assist people in remembering words. For example: economies of economy ([ ɪ 'k ɔ n ə mi ] depends on the farmer), pregnant pregnancy ([' pregron ə nsi ]

Male), pest ([ pest ] claps it). The memory of many words can be assisted by making one pronunciation similar to its pronunciation into another pronunciation or a similar sentence, and by using similar pronunciation and similar semantics to help people to memorize better. The word generated by this method may be referred to as an associative word of the word, also referred to as a white-ear or harmonic word of the word. Only a small fraction of the words currently being addressed by people, such as those listed above for pregnant pregnancy ("pregron ə nsi") are designed with associated memory-aid pronunciations. However, no one has proposed a better method for automatically generating or synthesizing the empty ears or harmonic words of any word. The present invention can generate harmonic sound words similar to Chinese pronunciation for any word. The English automation for generating Chinese association words is realized. The word memory device helps people to better memorize words.

Disclosure of Invention

the invention provides an associative word generating method and device based on word vectors, which are used for generating Chinese associative words corresponding to words when the words are memorized.

The invention provides a method for generating associative words based on word vectors, which mainly comprises the following steps:

acquiring a word to be subjected to harmonic sound generation, wherein the word contains a phonetic symbol;

Decomposing the phonetic symbols into syllables, wherein the syllables are decomposed according to the number of vowels and are divided into syllables;

inquiring an initial and final combination table according to syllable pronunciations as prefixes, acquiring pinyins containing the syllable pronunciations and acquiring Chinese characters corresponding to the pinyins;

aiming at the Chinese characters, combining the combined words, verifying the combined words and obtaining candidate words;

Acquiring the part of speech of the candidate word, and preferentially selecting the candidate word according to the part of speech;

Acquiring collocation candidate words according to a word vector technology, and sequencing collocation similarity according to similarity between words;

and acquiring word similarity between the English word Chinese explanation and the candidate words, and reordering the collocation similarity to finally obtain the associated words of the words.

Further optionally, in the method as described above, the decomposing the phonetic symbols into syllables mainly includes:

And counting the number of vowels in the phonetic symbols. The vowels are classified into monosyllabic, bisyllabic and polysyllabic according to whether they are two vowels or multiple vowels.

when there is only one consonant between two vowels, the consonant is divided into the following syllables.

when a plurality of consonant letters exist between two vowels, the two consonant letters are divided into front and back syllables respectively.

Further optionally, in the method as described above, the querying an initial/final combination table according to the pronunciation of the syllable as a prefix, obtaining a pinyin including the pronunciation of the syllable, and obtaining a chinese character corresponding to the pinyin, mainly includes:

Corresponding each syllable to the same pronunciation in the pinyin, taking the syllable as the pronunciation prefix of the syllable, and searching and matching on the initial and final combination table; and taking the matching result as candidate pinyin.

wherein, the obtaining of the Chinese characters corresponding to the syllables and the pinyin comprises the following steps: and screening and matching the characters with the pronunciation prefixes in the dictionary through the pronunciation prefixes, and taking the matched characters as candidate pronunciation characters of the vowel.

further optionally, in the method as described above, the synthesizing a combined word for the chinese character, verifying the combined word, and obtaining a candidate word mainly includes:

and sequentially combining every two words in front and back according to the candidate pronunciation characters to obtain combined words. If the front and the back characters can be combined into a correct Chinese phrase, the Chinese phrase is obtained;

If the front and back characters can not be combined into a correct Chinese phrase, the pinyin is used for similarity calculation to obtain the phrase.

further optionally, in the method as described above, the obtaining of the phrase by performing similarity calculation with pinyin mainly includes:

And calculating the similarity of the combined word and the pinyin of the word list, and acquiring n words with the highest pinyin similarity as the pronunciation words. In the combination process, some voices can be matched without words, and can be skipped to form single words by self.

further optionally, in the method, the obtaining part of speech of the candidate word and preferentially selecting the candidate word according to the part of speech mainly includes:

and performing part-of-speech tagging on the words to obtain the part-of-speech of the words and selecting the words with more entity images. Including graphics, objects, action words, and nouns and verbs as more preferable words.

Further optionally, in the method described above, the obtaining matching candidate words according to a word vector technique, similarity between words, and ranking the matching similarity mainly includes:

and performing semantic similarity calculation on each candidate word through word2vec word vector technology. And obtaining semantic association between every two, and sequencing the association.

further optionally, in the method as described above, the obtaining of the chinese explanation of the english word and the word similarity between the candidate words and reordering the collocation similarities includes:

and acquiring word translation, and calculating the semantic relevance of the translation words and the list. And reordering the list to obtain the first word with the most similar semantics as the associated word of the word.

the invention provides an associative word generating device based on word vectors, which mainly comprises:

the acquisition module is used for acquiring a word to be generated by the associated word and decomposing the word into syllables according to the phonetic symbol of the word;

The first generation module is used for taking syllable pronunciation as a prefix, looking up a table to obtain the pinyin of the syllable and obtaining a Chinese character corresponding to the pinyin;

the second generation module is used for synthesizing the combined words and verifying the combined words according to the Chinese characters by the user to obtain candidate words;

The selection module is used for calculating word relevance between the candidate words according to the parts of speech of the candidate words, a word vector technology and word Chinese explanation, and selecting the most appropriate associated words;

The technical scheme provided by the embodiment of the invention has the following beneficial effects:

When people memorize English words, the Chinese associative words corresponding to the words are generated to help memorize. The method is accurate and effective in generating the association words, considers the relation of each dimension such as voice collocation, translation collocation, voice collocation and the like, can be more efficient than the manual design of the association words, and has good effect.

drawings

FIG. 1 is a flow chart of an embodiment of a method for generating associative words of the present invention;

FIG. 2 is a block diagram of an embodiment of an association word generating apparatus according to the present invention;

FIG. 3 is a schematic diagram of matching of initials and finals according to the present invention.

Detailed Description

In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in detail with reference to the accompanying drawings and specific embodiments.

fig. 1 is a flowchart of an embodiment of a method for generating an associative word based on a word vector according to the present invention. As shown in fig. 1, the method for generating an associative word based on a word vector in this embodiment may specifically include the following steps:

Step 101, obtaining a word and obtaining the phonetic symbol of the word. The phonetic symbols may be obtained from a word dictionary or related software, such as a dictionary with channels. And decomposing the phonetic symbols into syllables.

according to the English international phonetic symbols, the total number of vowels is 20, and the number of consonants is 28.

The number of Chinese characters to be combined with the pronunciation is determined according to the number of syllables. The word phonetic symbol is decomposed into syllables, and the number of vowels in the phonetic symbol is mainly counted. The vowels are classified into monosyllabic, bisyllabic and polysyllabic according to whether they are two vowels or multiple vowels.

According to the english pronunciation convention, when there is only one consonant between two vowels, this consonant is divided into the following syllables.

By this method, the phonetic symbols of a word can be divided into different syllables. For example:

economy [ɪ'kɒnəmi]

by the above rule, there are four vowels in total. Belonging to multi-syllable words. According to the previous method, the word is decomposed into ɪ', k ɒ, n ə, mi four syllables. It can be understood that the word harmonious sound' built by people now also consists of four words, and it can be seen that the word is decomposed into syllables, which is beneficial to the initial construction of the number of words of the word composing the Chinese character.

and 102, determining the corresponding pinyin pronunciation prefix of each syllable. The pinyin pronunciation prefix refers to pronunciation formed by matching initial consonants or vowels of the syllable or initial consonants and vowels of the syllable.

to obtain the pinyin pronunciation prefix, each syllable and the corresponding same pinyin pronunciation can be searched and matched in the initial and final matching table, and the matched pinyin is used as the pinyin pronunciation prefix of the syllable.

for example, in step 101, after the word economy is decomposed, the first pronunciation is the I pronunciation, and the table is shown in fig. 3 by searching the initial and final tables of pinyin. The method can search and match yi, which belongs to the same pronunciation, and then can further match di, ti, ni, li, bi, pi, mi and other pronunciations also carry the I sound, so that di, ti, ni, li, bi, pi, mi can be used as the pronunciation prefix of the I pronunciation. The matching chart of the initial consonants and the simple or compound vowels is a tool for learning pinyin, and a more complete version can be obtained on the internet.

And 103, screening and matching characters with the pinyin pronunciation prefix according to the pinyin pronunciation prefix and corresponding to a dictionary. The character is used as a candidate pronunciation character of the vowel. For example, according to the above example, sound I in economies can be selected to have a word such as clothing, meaning, past, or "Yi", sound k ɒ can be selected to have a word such as "mouth, near, spread, or look", which are all words including the vowel. The dictionary for screening or query can adopt a Xinhua dictionary, and can also be a Chinese dictionary query system of an electronic version called through a network interface.

And 104, combining words aiming at the character groups, wherein the words can not be combined and are obtained through the pinyin similarity. The method mainly comprises the steps of obtaining a Chinese word list, wherein the Chinese word list can be called by a network interface by adopting an electronic version of a modern Chinese dictionary. The quantity of the Chinese common words is about more than 20 ten thousand words, and almost most of the words can be covered. Then, the word list is subjected to pinyin conversion; for example, the words farmers are converted into nongmin, and the whole dictionary can be converted into pinyin by means of yikao.

and combining all the words generated in sequence in pairs. For example, economi may be converted to four-word pinyin concatenations via the previous process, provided that the concatenated pronunciations are such that: yikaonongmin, then two groups will be yikao, kaong, nongmin, by matching the pinyin in the dictionary, it can be found that there are two words yikao, nongmin that can be matched, i.e. they can be combined into one word; if the front and back characters can be combined into a Chinese phrase, the word is obtained, i.e. the matching of the two characters can be converted into the harmonic sound of the two characters by using the word.

All words, if any, cannot be combined into words. These words are converted into pinyin. And calculating the similarity of the pinyin pairs with the word list, and acquiring n words with the highest pinyin similarity as the pronouncing words. In the combination process, some voices can be matched without words, and can be skipped to form single words by self. For example, it is assumed that the pronunciation of a word obtained by economi may also be yikuangnuomi by the above method, and yikuang does not have a word whose pinyin is the same as it and cannot be matched with a word in a dictionary, so that a word with the highest similarity to the pinyin, such as yikuan, can be calculated for the pinyin and converted into a word of 'one style'. Such words are also memorable.

and 105, acquiring the part of speech of the word, and selecting the word with a more entity image. Because the figures, objects or actions are easier to be memorized by people than the adjectives of the fictional words and the like, the parts of speech are labeled, the parts of speech are obtained, and nouns and verbs are extracted as preferred words.

for example, in the process of converting a plurality of characters into words, every two characters can be pronounced to form a plurality of different words, and the words need to be ordered. Then the noun and verb are obtained as the better word. For example, the conversion of pest to 'pat' is better than 'Person' because pat is easier to remember, it belongs to a verb, and it is a figure. Therefore, noun verbs are arranged in front of the noun verbs for screening harmonic words.

and 106, acquiring the semantic association degree of word collocation according to the word vector of the word or the word, and realizing the optimal collocation. And performing semantic similarity calculation on each candidate word through word2vec word vector technology. And obtaining the semantic association degree between every two. The relevance is ranked. For example, economi, the word, after the conversion of the previous steps, can obtain 'dependent, farmer', one, farmer ', art test, glutinous rice' and the like, different compound words. At this time, if the words are converted into word vectors, the direct connection similarity of the words can be calculated, and the matching of 'depending on' peasants 'can be easily obtained according to the correlation degree between semantics, so that the two words can be further correlated, and the word vectors of the' artistic investigation 'and the' glutinous rice 'are far away, so that the matching is not good depending on the peasants' and should be arranged behind the 'depending on the peasants'.

And step 107, acquiring the text similarity of the synthesized word and the word translation. The association is enhanced.

And acquiring word translation, and calculating the semantic relevance of the translation words and the list. The list is reordered. And acquiring the first word with the most similar semantic meaning as the associated word of the word.

for example, the word economi is translated into 'economy', then the word more related to it, the word of farmer is more semantically related than the word of glutinous rice, so that the 'farmer' is still in front of the 'glutinous rice'. The semantic relevance can still be calculated by a word vector technology or a synonym forest, a known network dictionary and the like.

through the above-mentioned way of preferably ordering words from tone to word, from word to word, and then multiple times, the word with the top rank and the best rank is finally selected as the associative word or harmonic word for memorizing the word.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. A method for generating an associative word based on a word vector, the method comprising:

2. The method of claim 1, wherein the decomposing the phonetic symbols into syllables comprises:

And counting the number of vowels in the phonetic symbols. Dividing the vowels into monosyllabic, bi-syllables and multi-syllables according to whether one vowel is two vowels or a plurality of vowels;

When only one consonant letter exists between two vowels, dividing the consonant letter into the following syllables;

3. The method as claimed in claim 1, wherein the querying an initial and final combination table according to the pronunciation of the syllable as a prefix, obtaining the pinyin including the pronunciation of the syllable, and obtaining the chinese character corresponding to the pinyin, mainly comprises:

Corresponding each syllable to the same pronunciation in the pinyin, taking the syllable as the pronunciation prefix of the syllable, and searching and matching on the initial and final combination table; taking the matching result as a candidate pinyin;

4. The method of claim 1, wherein the synthesizing a combined word for the Chinese character, verifying the combined word, and obtaining a candidate word mainly comprises:

5. the method of claim 4, wherein the similarity calculation using pinyin to obtain a phrase mainly includes:

6. The method of claim 1, wherein the obtaining of the part of speech of the candidate word and the preferentially selecting of the candidate word according to the part of speech mainly comprise:

and performing part-of-speech tagging on the words, acquiring the part-of-speech of the words, selecting words with more entity images, including figures, objects and action words, and taking nouns and verbs as more preferable words.

7. The method of claim 1, wherein the obtaining collocation candidate words according to a word vector technique, similarity between words, and ranking collocation similarities mainly comprises:

and performing semantic similarity calculation on each candidate word through word2vec word vector technology. And obtaining the semantic association degree between every two. The relevance is ranked.

8. the method of claim 1, wherein the obtaining of the word similarity between the chinese explanation of the english word and the candidate word and the reordering of the collocation similarity comprises:

9. An apparatus for generating suggested words based on word vectors, the apparatus comprising:

And the selecting module is used for calculating word relevance between the candidate words according to the parts of speech of the candidate words, the word vector technology and the word Chinese explanation, and selecting the most appropriate associated words.