TWI681304B - System and method for adaptively adjusting related search words - Google Patents

Abstract

A system for adaptively adjusting related search words includes an input device, a search log collection module, a threshold setting module, and a process evolution module. The input device is configured to receive a search term. The search log collection module is configured to determine whether a cumulative number of search words in historical search logs is greater than a first threshold or less than a second threshold. The threshold setting module is configured to set the first and second threshold in terms of the number of search words in the search logs . When the cumulative number of search words is between the first threshold and the second threshold, the process evolution module further optimizes the middle search process to find out at least one related word in the indexed text and/or at least one historical search word in the historical search logs that are most relevant to the attributes or content of the search word.

Description

System and method for adaptively adjusting related search words

The invention relates to a system and method for adaptively adjusting related search words.

Modern search systems usually give back to users other search terms related to the search term in the search results, to help users quickly clarify the query target, because the search keywords used by users often cannot accurately describe them in short vocabulary The search intention, or the user's given search term or search target has multiple descriptions or ambiguities that cause the vocabulary between the user and the text to not match, or the user's understanding or knowledge of the search target is insufficient and misused Search terms, or user typing errors such as sound or near sound and other factors. In general, the related search term extraction technology can be divided into a method based on index text content and a method based on historical query records according to the data source. The text-based method is used in the early stage of the search system, and it can immediately provide a list of related search terms based on the relevant analysis of the vocabulary in the index text content, but its disadvantage is that it can only provide suggestions based on the fixed text content, and cannot be based on the later stage. The accumulated historical query records analyze and predict the user's search intentions. Although the method based on historical query records can provide the latest search intent prediction based on the accumulated user data, so as to obtain a better list of related search terms, but it cannot provide suggestions immediately in the early stage of the system, and it takes a long time for users Use only in order to accumulate enough analysis data sources. In the conventional method, the weight integration method is also used to combine the above two methods, so that whether there is no user historical data in the early stage of the search system or when there is sufficient historical data in the later stage, related search terms can be recommended.

However, the weight integration method also has the problem of weighted combination of data sources. Manual setting often fails to achieve the best results. Usually, it is necessary to accumulate enough search record data to train the first set of optimal weight combinations by statistical model or machine learning. And there are still difficulties in transfer learning in different vertical fields. Therefore, the above-mentioned extraction techniques are respectively applicable to search systems in different online periods. Due to the different search records, it is not possible to provide related search terms suitable for suggesting users at any time.

The invention relates to a system and method for adaptively adjusting related search words, which can self-adjust related search words according to the number of search records accumulated by the system to provide related search words suitable for suggesting users.

According to one aspect of the present invention, a system for adaptively adjusting related search terms is provided, including an input device, a record collection module, a threshold setting module, and an evolution module. The input device is used to receive user input and generate a search term. The record collection module is used to determine whether the accumulated search times of the search term is greater than a first threshold or less than a second threshold. The threshold setting module is used to set the number of search records that satisfy the first or second threshold. The evolution module is used to adjust a search process according to the number of search records. When the cumulative number of search terms for the search term is greater than the first threshold, the evolution module finds the content or attributes related to the search term based on a historical search record At least one historical search term. When the cumulative number of search terms for the search term is less than the second threshold, the evolution module performs an initial search process to find at least one related word in a text related to the content or attribute of the search term. When the cumulative number of search terms for the search term is between the first threshold and the second threshold, the evolution module optimizes the mid-term search process to further find the content or attributes of the search term in the text and historical search records At least one related word and/or at least one historical search word that maximizes relevance.

According to one aspect of the present invention, a method for adaptively adjusting related search terms is proposed, including the following steps. The input process is used to receive user input and generate a search term. The record collection process is used to determine whether the cumulative number of search terms for a search term is greater than a first threshold or less than a second threshold. The threshold setting process is used to set the number of search records that satisfy the first or second threshold. The evolution process is used to adjust a search process according to the number of search records. When the cumulative number of search terms for the search term is greater than the first threshold, the evolution process finds at least the content or attribute related to the search term based on a historical search record A historical search term. When the cumulative search frequency of the search term is less than the second threshold, the evolution process performs an initial search process to find at least one related word in a text related to the content or attribute of the search term. When the cumulative number of search terms for the search term is between the first threshold and the second threshold, the evolution process optimizes the mid-term search process to further find the content or attributes related to the search term in the text and historical search records Maximized at least one related word and/or at least one historical search word.

In order to have a better understanding of the above and other aspects of the present invention, the following examples are specifically described in conjunction with the accompanying drawings as follows:

The following is an example for detailed description. The example is only used as an example, not intended to limit the scope of the present invention. The following description uses the same/similar symbols to indicate the same/similar components. Directional terms mentioned in the following embodiments, for example: up, down, left, right, front or back, etc., are only directions referring to the drawings. Therefore, the directional terminology is used to illustrate rather than limit the invention.

According to an embodiment of the present invention, a system for adaptively adjusting related search terms is provided, such as a search engine with a self-adjusting search process. For the search engine imported into the system in the early stage, before a sufficient number of search records have been accumulated, the system can compare the text content or characteristics of the search word in the text based on the indexed text and the index vocabulary in the initial stage At least one related word related to the attribute to establish an initial related search word list. Then, after accumulating a certain number of search records in the mid-term, the system can maximize the correlation between the content and attributes of the search term in the text and the historical search records based on a certain number of historical search records and texts that have been indexed in the early stage At least one related word and/or at least one historical search word to create a mid-term related search word list. After that, after accumulating a sufficient number of search records in the later period, the system can directly find at least one historical search term related to the content or attribute of the search term based on the search term input by the user, to establish a related search term list in the later period.

As can be seen from the above, the system can achieve self-optimization according to the number of search records accumulated in different periods, so that its evolution module can smoothly evolve from the stage of no user behavior records (search records) in the early stage to the later user behavior records (Search history) Main stage, and then provide related search terms suitable for suggesting users.

Referring to FIG. 1, according to an embodiment of the present invention, a system 100 for adaptively adjusting related search terms includes an input device 110, a record collection module 120, a threshold setting module 130, and an evolution module 140 . The input device 110 is used to receive user input and generate a search term 112. The record collection module 120 is used to determine whether the accumulated search times of the search term 112 is greater than a first threshold value or less than a second threshold value (represented by the threshold value 132). The threshold setting module 130 is used to set the number of search records that satisfy the first or second threshold. In addition, the evolution module 140 is used to adjust a search process according to the number of search records.

In one embodiment, the input device 110 may be a user interface for reading user input data, including text, symbols, and/or voice. Taking a computer or a remote server as an example, the input device 110 may be a handheld electronic device connected to the computer or the remote server. The present invention is not limited to this. The input device 110 may input the search term 112 to be searched by the user to In a computer or a remote server, the search engine 102 imported into the system 100 is used to search for data in an online or local text database. The database may include a record database 124 and a text database 126. The text database 126 is used to store the source of the text 114 to be searched, including text files and/or database fields: text files such as product manual files, advertisement copy files, product test report files, web page files, etc.; database fields For example, the data fields of the product database, such as the product name, keywords, product description, brand, etc. The record database 124 is used to store the user's historical search records 126.

The record collection module 120 is used to collect user operations on the system 100, including inputting search terms, click locations, clicks, browsing time and other information, as well as the content or attributes of each search term 112. After the record collection module 120 collects the above data, it becomes the historical search record 126 and further stores it in the record database 124. The content or attribute of the search word 112 may be the product's Chinese name, English name, abbreviation, brand, model, function, and other brand names, etc. The present invention is not limited to this. The content or attribute of the search word 112 may be based on the dictionary The meanings in the words or user-defined semantics or manually edited open data (such as Wikipedia, DBpedia, Open Directory Project) or statistical proper noun recognition (Name Entity Recognition) are determined. After the content or attribute of the search word 112 is determined, the system 100 then searches for related related words 148 according to the content or attribute of the search word 112.

In addition, the system 100 can also use the search engine 102 to parse and grammatically reconstruct the search term 112, and filter words in the text 114 and/or historical search records 126 that are not related to the content or attributes of the search term 112 to ensure data retrieval. Correctness and continuity.

In addition, the threshold setting module 130 is used to set the number of search records that satisfy the first or second threshold. The number of search records is not limited to the accumulated number of search words 112 with only the same vocabulary, but may also be the accumulated number of search words 112 of the same type with different vocabularies but similar semantic meanings. When different users search for the same type of search terms 112 or similar search terms 112, the system 100 can accumulate or perform weighting on the search records of the same type or similar search terms 112. When the system 100 accumulates the search records When the number reaches a threshold 132, the evolution module 140 of the system 100 adjusts the search process adaptively according to the number of search records, as shown in Figures 2, 3, and 4.

Referring to FIG. 1, the system 100 may further include a word segmentation module 146, a record related word generation module 160 and a text related word generation module 150. The index vocabulary 144 includes a set of word lists, and each string can be composed of one or more alphanumeric characters or symbols. The index vocabulary can be pre-set manually, or a general general dictionary or a professional dictionary, or by word segmentation After analyzing the content of the text 114, the module 146 aggregates all the string phrases to form the index vocabulary 144, or it may be a combination of the foregoing methods, for example, combining all the words analyzed by the word segmentation module 146 by combining the professional field dictionary and the text. The content of the text 114 can be a specified data table or data field of a document, web page, or database. For example, if the target of the search system is a product, the content of the text 114 can be the product name and product description of the product data table in the product database , Product keywords and other database fields, as well as product description webpage content.

The word-breaking module 146 can divide the search words 112 (for example, Chinese words) input by the user into meaningful phrases. For example, the search word 112 input by the user is a chip card reader, and the word breaker module 146 may divide the chip reader into a chip and a card reader, or only a card reader. Therefore, when the search word 112 does not exist in the text 114, the word breaker module 146 performs byte analysis, word analysis, or word comparison according to the index word table 144, and disassembles the search word 112 into at least one index word. For the search engine 102 to further search for the index words appearing in the text 114. The above-mentioned Chinese words can use dictionary-based word breaking algorithm, forward maximum matching algorithm, reverse maximum matching algorithm or bidirectional maximum matching algorithm, or a corpus-based statistical word breaking algorithm such as Conditional Random Fields (CRF) ) Or deep neural networks (Deep Neural Networks, DNN), etc., the present invention is not limited to this.

In addition, the text related word generation module 150 can analyze the first M index words in the text 114 that are most relevant to the search word 112 according to the index word list 144 to generate a text related word list 152. M is, for example, 5 or more positive integers. As described above, in one embodiment, the text related word generation module 150 can calculate a related strength by the probability that the search word 112 and the index word appear alone or co-exist in the text 114. The stronger the related strength, the greater the relatedness. The stronger, conversely, the weaker the connection strength, the worse the connection. The calculation of the above-mentioned correlation strength can be achieved by the correlation learning method, Pointwise Mutual Information (PMI), PMI improved algorithm, KL divergence algorithm (Kullback–Leibler divergence), standardized Google distance algorithm Method and algorithm based on Wordnet distance, the invention is not limited to this.

In addition, the record related word generation module 160 is used to analyze the degree of connection between any two historical search words in the historical search record 122 and find the top N historical search words most relevant to the search word 112 to generate a record Related words table 162. N is, for example, 5 or more positive integers. As described above, in one embodiment, the record related word generation module 160 may calculate a related strength by the probability that the content or attribute of the current search word 112 and the historical search word appear alone or co-exist in the historical search record 122, The stronger the connection strength, the stronger the connection. On the contrary, the weaker the connection strength, the worse the connection. In addition, in addition to comparing the occurrence position of the vocabulary content, the correlation degree can also be calculated according to other attributes of the search term in the historical search record 122, such as click position, number of clicks, browsing time and other attributes. For example, Pointwise Mutual Information (PMI) is used, but other algorithms can also be used, such as related learning method, PMI improved algorithm, KL divergence algorithm (Kullback–Leibler divergence), standardization The Google distance algorithm and the algorithm based on wordnet distance are used to achieve this, and the invention is not limited to this.

Please refer to FIG. 1. In order to optimize the mid-term search process, the system 100 further includes a related word discrimination calculation module 170 and a related word recommendation module 174. The related word discrimination calculation module 170 may calculate the discrimination value 172 of each related word 148 according to the text 114, the index word table 144, the recorded related word table 162, and the text related word table 152. The discrimination value 172 is used to judge the unique degree of the related word 148, that is, an index used to measure the difference degree of the related word 148 in the text 114. And it can be used to increase the diversity of the list of related words, to avoid the problem of too many similar related words. When the related words 148 only appear in a certain text 114, the higher the discrimination value; when the related words 148 appear in multiple texts 114 at the same time, the lower the discrimination value. For example, in multiple texts 114, the uniqueness of a related word 148 is inversely proportional to the frequency (document frequency, DF for short) of the related words 148 appearing in these texts 114, that is, the inverse document frequency ( inverse document frequency, referred to as IDF). Therefore, the related word discrimination calculation module 170 may use, for example, a reverse document frequency algorithm, a residual reverse document frequency (RIDF) algorithm, or a discrimination power algorithm. The present invention is not limited to this to calculate the correlation words 148. Discrimination value 172, and establish a matching table of related words 148 and discrimination value 172.

In one embodiment, when a certain related word 148 exists in the index word table 144, the related word discrimination calculation module 170 directly calculates the discrimination value of the index word. When a certain related word 148 does not exist in the index word list 144, after the word segmentation module 146 performs word segmentation of a certain related word 148, the related word discrimination degree calculation module 170 calculates the discrimination value for each index word after the word segmentation, Then, the discrimination values of the related words 148 are estimated by taking the minimum value, the maximum value, the arithmetic average value, or the weighted average value among the discrimination values.

In one embodiment, the system 100 further includes a new word recognition module 142 that can extract new words that are not included in the index word list from a given word. The calculation method of the new word recognition module 142 can be through language rules such as phonological rules or grammatical rules or word formation rules, or through statistical models such as Hidden Markov Model (HMM), conditional random field ( Conditional Random Fields (CRF), Support Vector Machine (SVM), Deep Neural Network (DNN), or through specific statistics such as Pointwise Mutual Information (PMI) algorithm Calculated in other ways. When a related word 148 does not exist in the index word list 144, the new word recognition module 142 extracts a part of the character string recognized as a new word from the related word 148, and then gives the discriminated value for evaluation. The calculation method may be a predetermined fixed value, or a dynamic weighted value based on the maximum value or the maximum value of the discrimination of all words in the index vocabulary 144. The non-new word part of the related word can continue to be calculated according to the index word table 144. If it exists in the index word table 144, the related word discrimination degree calculation module 170 directly calculates the discrimination value of the index word. Finally, the discriminant value of the new and non-new word partial phrases is obtained, and then the discriminant value of the related word 148 is estimated by taking the minimum value, or the maximum value, or the arithmetic average, or the weighted average value of these discriminating values. . If the part of the string of the non-new word does not exist in the index word list 144, the word segmentation module 146 performs word segmentation on the string to obtain at least one index word, and the related word discrimination calculation module 170 then targets each index after word segmentation Calculate the discrimination value of the word, and finally obtain the discrimination value of the new word and the non-new word part of the word group, and then use these minimum values, or the maximum value, or the arithmetic average, or the weighted average to estimate the relationship The discriminator value of word 148.

In addition, the related word recommendation module 174 is used to compare the discriminated value of each related word 148 in the recorded related word table 162 and the discriminated value of each related word 148 in the text related word table 152, and according to the order of the discriminated values of the related words 148, From the text related word list 152 and the recorded related word list 162, select the first P related words 148 with a higher discrimination value. P is, for example, 5 or more positive integers. In this way, a related search vocabulary 176 suitable for suggestions can be completed.

Please refer to FIGS. 1 and 2, wherein FIG. 2 is a schematic diagram of an initial search process performed by the system 100 for adaptively adjusting related search terms 176 according to an embodiment of the present invention, which includes steps S11-S14. Please refer to steps S11 and S12 to determine whether the search term 112 is in the search record, and if so, to further determine whether the cumulative search times of the search term 112 is less than the second threshold. When the above two conditions are met, the evolution module 140 performs an initial search process. At this time, since the search term 112 does not exist in the historical search record 122 or the cumulative number of searches for the search term 112 is very small, the search engine 102 cannot Of search terms 112 to find historical search terms that are suitable for suggestions. Please refer to steps S13 and S14 to determine whether the search word 112 is in a text 114. If not, the word breaker module 146 disassembles the search word 112 into at least one index word according to the index word table 144, and returns to step S11. Further determine whether the index word is in the search record. When the search word 112 exists in a text 114, the text related word generation module 150 can find at least one related word 148 in a text 114 related to the content or attribute of the search word 112 according to the built-in text 114 and the index word list 144 .

Next, please refer to FIGS. 1 and 3, wherein FIG. 3 shows a schematic diagram of the system 100 for adaptively adjusting related search terms according to an embodiment of the present invention to optimize the mid-term search process. The process steps of this embodiment are the same as the above embodiments, except that in step S12, when the cumulative search times of the search term 112 is greater than the second threshold and less than the first threshold, the system 100 accumulates a certain number of searches The record can be used by the evolution module 140 to perform an intermediate search process. At this time, the record related word generation module 160 may find at least one historical search word related to the content or attribute of the search word 112 according to a historical search record 122. Therefore, the search engine 102 can not only find suitable related words 148 according to the current search word 112, but also find suitable related words 148 according to the built-in text 114 and the index word list 144. The related word recognition degree calculation module 170 and the new word recognition module 142 generate the related word discrimination value, and then select through the related word recommendation module 174 to further find at least one that maximizes the correlation with the content or attribute of the search word 112 The related words 148 and/or at least one historical search word are used to obtain an optimized search related word list 176.

Next, please refer to FIGS. 1 and 4, wherein FIG. 4 illustrates a schematic diagram of the post-search process of the system 100 for adaptively adjusting related search terms according to an embodiment of the present invention, which omits steps S13 and S14 in the initial stage. For the text search process, only the judgment steps of steps S11 and S12 can be performed. In this embodiment, when the search term 112 appears in the historical search record 122, and the cumulative search times of the search term 112 is greater than the first threshold value and greater than the second threshold value, because the system 100 has accumulated a sufficient number of historical search records 122, so that the evolution module 140 can perform a later search process. At this time, the record related word generation module 160 may find at least one historical search word related to the content or attribute of the search word 112 according to a historical search record 122. Therefore, the search engine 102 does not need to find the related keywords 148 suitable for suggestions based on the built-in text 114 and the index vocabulary 144, but directly finds the related keywords suitable for suggestions from a historical search record 122 based on the current search word 112 148. The first threshold and the second threshold are the cumulative search times of the search term 112, which can be determined according to the concept of general statistical large sample size (the sample size is greater than 30), or based on the search system of the same field and similar scale, for example In the field of shopping search, based on the number of similar products, enough cumulative search times are needed to reach the user’s satisfaction for the record related words to set the first and second thresholds. Or, in the process of using the search system 100, domain experts can dynamically adjust the first and second thresholds according to the search results, to adjust the speed of the early stage of evolution to the late stage, or from the middle or late stage to the previous one. Stage.

In one embodiment, the method for adaptively adjusting the related search term 176 can be implemented as a software program, which can be stored in a non-transitory computer readable medium, such as a hard disk , CD, flash drive, memory and other program storage devices. When the processor loads this software program from a non-transitory computer readable medium, it can perform the method flow shown in Figures 2, 3, and 4 to perform an initial search. The process evolves into a mid-term search process, and then evolves from a mid-term search process to a late-stage search process.

In one embodiment, the system 100 for adaptively adjusting related search terms may include a processor and a program storage device. The processor can execute one or more computer-executable instructions, and the program storage device stores a computer program model executable by the processor. Group, in which the computer program module, when executed by the processor, causes the processor to perform the operations as shown in Figures 2, 3, and 4.

In another embodiment, the aforementioned record collection module 120, threshold setting module 130, evolution module 140, new word recognition module 142, text related word generation module 150, recorded related word generation module 160, The related word discrimination calculation module 170 and the related word recommendation module 174 may be individually implemented as software units or hardware units, or some modules may be combined and implemented by software, and some modules are combined and implemented by hardware. The module implemented by software can be regarded as an operation process, that is, record collection process, threshold setting process, evolution process, new word recognition process, text related word generation process, record related word generation process, related word discrimination degree calculation process, Related word recommendation process, etc., can be loaded by the processor to perform the corresponding function. A module implemented in hardware, for example, can be implemented as a microcontroller, a microprocessor, a digital signal processor, and an application specific integrated circuit (ASIC) , Digital logic circuit, or field programmable gate array (FPGA).

The system and method for adaptively adjusting related search terms disclosed in the above embodiments of the present invention can self-adjust related search terms according to the number of accumulated search records of the system to provide related search terms suitable for suggesting users, thereby reducing the system The manpower and time cost required for program development, and there is no need to learn the first set of weight combinations in advance, and there is no problem of vertical field conversion learning. In addition, the present invention also considers the situation that the search term recommendation process can continue to evolve as the search record changes, and establishes a higher accuracy search term recommendation mechanism, which can avoid a single search term recommendation process that may generate content related to the search term Or the problem of related words whose attributes are not related, increase the convenience of management and improve the flexibility of use.

In summary, although the present invention has been disclosed as above with examples, it is not intended to limit the present invention. Those with ordinary knowledge in the technical field to which the present invention belongs can make various modifications and retouching without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be deemed as defined by the scope of the attached patent application.

100‧‧‧Adaptive system for adjusting related search words 102‧‧‧ search engine 110‧‧‧Input device 112‧‧‧ search terms 114‧‧‧ text 120‧‧‧Record collection module 122‧‧‧History search record 124‧‧‧Record database 126‧‧‧ Text database 130‧‧‧ Threshold value setting module 132‧‧‧ Threshold 140‧‧‧Evolution module 142‧‧‧New word recognition module 144‧‧‧ Index word list 146‧‧‧ Word Breaking Module 148‧‧‧ related words 150‧‧‧ text related word generation module 152‧‧‧ List of related words 160‧‧‧Record related word generation module 162‧‧‧List of related words 170‧‧‧ related word discrimination calculation module 172‧‧‧discrimination value 174‧‧‧Related Links Recommendation Module 176‧‧‧ related search vocabulary

FIG. 1 is a schematic diagram of a system for adaptively adjusting related search terms according to an embodiment of the invention. FIG. 2 is a schematic diagram of an initial search process of a system for adaptively adjusting related search terms according to an embodiment of the invention. FIG. 3 is a schematic diagram of the system for adaptively adjusting related search terms according to an embodiment of the present invention to optimize the mid-term search process. FIG. 4 is a schematic diagram of the post-search process of the system for adaptively adjusting related search terms according to an embodiment of the invention.

100‧‧‧Adaptive system for adjusting related search words

102‧‧‧ search engine

110‧‧‧Input device

112‧‧‧ search terms

114‧‧‧ text

120‧‧‧Record collection module

122‧‧‧History search record

124‧‧‧Record database

126‧‧‧ Text database

130‧‧‧ Threshold value setting module

132‧‧‧ Threshold

140‧‧‧Evolution module

142‧‧‧New word recognition module

144‧‧‧ Index word list

146‧‧‧ Word Breaking Module

148‧‧‧ related words

150‧‧‧ text related word generation module

152‧‧‧ List of related words

160‧‧‧Record related word generation module

162‧‧‧List of related words

170‧‧‧ related word discrimination calculation module

172‧‧‧discrimination value

174‧‧‧Related Links Recommendation Module

176‧‧‧ related search vocabulary

Claims (18)

A system for adaptively adjusting related search terms, including: an input device for receiving a search term; a record collection module for determining whether the accumulated search times of the search term is greater than a first threshold or less than one A second threshold; a threshold setting module for setting the number of search records that satisfy the first or the second threshold; and an evolution module for adjusting a search process based on the number of search records , Wherein when the cumulative search times of the search term are greater than the first threshold, the evolution module finds at least one historical search term related to the content or attribute of the search term according to a historical search record, wherein, when the search When the cumulative number of search terms for a word is less than the second threshold, the evolution module performs an initial search process to find at least one related word in a text that is related to the content or attribute of the search term, where, when the search term When the cumulative number of searches is between the first threshold and the second threshold, the evolution module optimizes the mid-term search process to further find the search terms in the text and in the historical search records The at least one related word and/or the at least one historical search word that maximize content or attribute relevance. The system as described in item 1 of the patent application scope further includes: an index vocabulary; and a text related word generation module, which is used to analyze the top M in the text most relevant to the search term according to the index vocabulary Index words to generate a list of text related words; and a record related word generation module to analyze the degree of connection between any two historical search words in the historical search record to find the most relevant to the search word The first N historical search words to generate a list of related words of record. The system as described in item 2 of the patent application scope, wherein the text related word generation module calculates a related strength based on the probability that the search word and the index words appear alone or together in the text. The system as described in item 2 of the patent application scope, wherein the record related word generation module calculates a connection based on the probability that the search term and the content or attributes of the historical search terms appear alone or together in the historical search record strength. The system as described in item 2 of the patent application scope further includes: a related word discrimination calculation module, which calculates the discrimination value of each related word based on the index word list, the record related word list and the text related word list; And a related word recommendation module, used to compare the discriminated value of each related word in the recorded related word list with the discriminated value of each related word in the text related word list, and sorted according to the discriminated value of each related word, Select the top P related words with higher discrimination value from the text related word list and the recorded related word list. The system as described in item 5 of the patent application scope, wherein the related word discrimination degree calculation module calculates the discriminant value according to a degree of difference in which each related word appears in the text, and the difference degree and each related word appear in a single The text or frequencies in the text are related. The system as described in item 2 of the patent application scope further includes: a segmentation module for receiving the search term, when the search term does not exist in the text, the segmentation module according to the index word list will The search term is disassembled into at least one index term. The system as described in item 5 of the patent application scope further includes: a new word recognition module for identifying whether the related word contains a new word that does not exist in the index vocabulary, wherein when the related word contains the new word When a word is used, the related word discrimination degree calculation module calculates the discrimination value of the related word based on the related word and the included new word. The system as described in one of claims 1 to 8, wherein the system is executed by a processor or a software program loaded by the processor. A method for adaptively adjusting related search terms, including: an input process for receiving a search term; a record collection process for determining whether the cumulative search times of the search term is greater than a first threshold or less than a first Two thresholds; a threshold setting process for setting the number of search records that satisfy the first or the second threshold; and an evolution process for adjusting a search process based on the number of search records When the cumulative search times of the search term are greater than the first threshold, the evolution process finds at least one historical search term related to the content or attribute of the search term according to a historical search record, wherein, when the cumulative search of the search term When the number of times is less than the second threshold, the evolutionary process performs an initial search process to find at least one related word in a text that is related to the content or attribute of the search term. Between the first threshold and the second threshold, the evolutionary process optimizes the mid-term search process to further find the text and the historical search record to maximize the content or attribute related to the search term The at least one related word and/or the at least one historical search word. The method described in item 10 of the patent application scope further includes: creating an index vocabulary; and a text related word generation process, based on the index vocabulary, to analyze the top M most relevant to the search term in the text Index words to generate a list of text related words; and a record related word generation process to analyze the degree of connection between any two historical search words in the historical search record to find the most relevant front N historical search words to generate a list of related words. The method as described in item 11 of the patent application scope, wherein the text related word generation process calculates a related strength based on the probability that the search word and the index words appear alone or together in the text. The method as described in item 11 of the patent application scope, wherein the record related word generation process calculates a related strength based on the probability that the search terms and the contents or attributes of the historical search terms appear alone or together in the historical search records. The method as described in item 11 of the patent application scope further includes: a process of calculating the degree of discrimination of related words, calculating the identification value of each related word based on the index word list, the recorded related word list and the text related word list; and A related word recommendation process for comparing the discriminating value of each related word in the related word list of the record with the discriminating value of each related word in the text related word list, and according to the order of the discriminating value of each related word, from the The first P related words with higher identification value are selected in the text related word list and the related word list of the record. The method as described in item 14 of the patent application scope, wherein the related word discrimination degree calculation process calculates a discriminant value according to a degree of difference in which each related word appears in the text, and the difference degree and each related word appear in a single The text or frequencies in the text are related. The method as described in item 11 of the patent application scope further includes: a word segmentation process to receive the search term, and when the search term does not exist in the text, the word segmentation process will search according to the index vocabulary The word is disassembled into at least one index word. The method as described in item 14 of the patent application scope further includes: a new word recognition process to identify whether the related word contains a new word that does not exist in the index word list, wherein when the related word contains the new word At this time, the related word discrimination degree calculation process calculates the related word discrimination value based on the related word and the included new word. The method according to one of the items 11 to 17 of the patent application scope, wherein the method is executed by a processor or a software program loaded by the processor.

Images