Patent document pre-evaluation method and device

Abstract

The present invention discloses a patent document pre-evaluation method and device, wherein the method comprises: inputting the patent document information to be evaluated into a pre-established semantic analysis model, retrieving multiple similar patent documents similar to the patent document to be evaluated; using each claim item of the patent document to be evaluated to traverse each similar patent document for feature comparison, and obtaining the technical feature similarity between each claim item of the patent document to be evaluated and each similar patent document; drawing a technical feature spectrum diagram according to the similarity; the vertical axis in the spectrum diagram corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document corresponds to the technical feature similarity, and the size of the similarity is corresponding to a preset color tone; and pre-evaluating the patent document to be evaluated according to the technical feature spectrum diagram. The present invention can pre-evaluate patent documents efficiently, accurately and intuitively.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a patent document pre-evaluation method and device. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] At present, the existing patent document pre-evaluation method is to manually compare the similarity of two documents to obtain the pre-evaluation result of the patent document to be evaluated. Therefore, the existing patent document pre-evaluation method is inefficient and less intuitive. Summary of the invention

[0004] The embodiment of the present invention provides a patent document pre-evaluation method for efficiently, accurately and intuitively pre-evaluating the patent document. The method includes:

[0005] Input the information of the patent document to be evaluated into a pre-established semantic analysis model to retrieve multiple similar patent documents similar to the patent document to be evaluated;

[0006] Using each claim item of the patent document to be evaluated to traverse each similar patent document to compare technical features, the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document is obtained;

[0007] A technical feature spectrum is drawn according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim item and each similar patent document, and the size of the similarity is corresponding to a preset color tone;

[0008] According to the technical feature spectrum, a preliminary evaluation is performed on the patent document to be evaluated to obtain a preliminary evaluation result of the patent document to be evaluated.

[0009] The embodiment of the present invention further provides a patent document pre-evaluation device for efficiently, accurately and intuitively pre-evaluating patent documents, the device comprising:

[0010] An intelligent retrieval unit, used to input the information of the patent document to be evaluated into a pre-established semantic analysis model, and retrieve multiple similar patent documents similar to the patent document to be evaluated;

[0011] A traversal feature comparison unit is used to traverse each similar patent document using each claim item of the patent document to be evaluated to compare the technical features, and obtain the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document;

[0012] A spectrum drawing unit is used to draw a technical feature spectrum according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim item and each similar patent document, and the size of the similarity is corresponding to a preset color tone;

[0013] The visual interactive evaluation unit is used to pre-evaluate the patent document to be evaluated according to the technical feature spectrum diagram to obtain a pre-evaluation result of the patent document to be evaluated.

[0014] An embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the pre-evaluation method of the above patent document when executing the computer program.

[0015] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the pre-evaluation method of the above patent document is implemented.

[0016] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the pre-evaluation method of the above patent document is implemented.

[0017] Compared with the prior art technical solution of pre-evaluating the patent documents to be evaluated by manually comparing the similarity of the documents, which is inefficient and less intuitive, the pre-evaluation solution for patent documents provided by the embodiment of the present invention has the following beneficial technical effects:

[0018] First, the embodiment of the present invention inputs the information of the patent document to be evaluated into a pre-established semantic analysis model, retrieves multiple similar patent documents similar to the patent document to be evaluated, realizes efficient retrieval through semantic analysis, and lays a solid foundation for the subsequent efficient preliminary evaluation of the patent document.

[0019] Secondly, the embodiment of the present invention uses a traversal algorithm to traverse each claim of the patent document to be evaluated through each similar patent document for technical feature comparison, and obtains the similarity of technical features of each claim of the patent document to be evaluated and each similar patent document, thereby realizing the use of the traversal algorithm for technical feature comparison: the patent document to be evaluated traverses all claims, and the similar patent document traverses all document contents. This algorithm logic is close to the business logic used by users, so that patent documents can be accurately pre-evaluated.

[0020] Again, in an embodiment of the present invention, a technical feature spectrum is drawn according to the similarity of technical features between each claim of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim and each similar patent document, and the size of the similarity is corresponding to a preset color tone; based on the technical feature spectrum, the patent document to be evaluated is pre-evaluated to obtain a pre-evaluation result of the patent document to be evaluated, thereby realizing a technical feature spectrum drawn based on the technical feature similarity, and performing visual interaction, so that the pre-evaluation result of the patent document to be evaluated can be intuitively obtained.

[0021] In summary, the patent document pre-evaluation solution provided by the embodiment of the present invention can pre-evaluate patent documents efficiently, accurately and intuitively. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0023] Figure 1 Schematic diagram of the process of pre-evaluating a patent document in an embodiment of the present invention;

[0024] Figure 2 It is a schematic diagram of the process of traversing feature comparison in an embodiment of the present invention;

[0025] Figure 3 It is a technical characteristic spectrum diagram in the embodiment of the present invention;

[0026] Figure 4 Schematic diagram of the principle of pre-evaluation of patent documents in an embodiment of the present invention;

[0027] Figure 5It is a schematic diagram of the principle of pre-evaluation of patent documents in another embodiment of the present invention;

[0028] Figure 6 An exemplary schematic diagram of the preliminary evaluation of patent documents in an embodiment of the present invention;

[0029] Figure 7 Schematic diagram of the structure of the pre-evaluation device of the patent document in the embodiment of the present invention. DETAILED DESCRIPTION

[0030] To make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the embodiments of the present invention are further described in detail below in conjunction with the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0031] The present invention proposes a pre-evaluation scheme for patent documents, which displays similar documents of a certain patent document in the form of a spectrum graph. The higher the similarity of the document, the greater the thermal value of the corresponding point on the spectrum graph (the heavier the red). The claims of a certain patent document (target patent document) are compared with the full texts of multiple similar patent documents. The comparison results are displayed in the form of a spectrum graph. Each square (intersection) on the spectrum corresponds to the target patent document (patent document to be evaluated) and a similar document (approximate patent document). Clicking a square can display the comparison results of the target patent document and a similar document; clicking a row can display the comparison results of a certain feature of the target patent document and each similar document, and clicking a column can display the comparison results of each feature in a similar document and the claims of the target patent document. According to the color of the spectrum graph, the novelty of the target patent document can be intuitively reflected. The pre-evaluation scheme of the patent document is introduced in detail below.

[0032] Figure 1 Schematic diagram of the process of pre-evaluation of patent documents in an embodiment of the present invention. Figure 1 As shown, the method comprises the following steps:

[0033] Step 101: inputting the information of the patent document to be evaluated into a pre-established semantic analysis model to retrieve a plurality of similar patent documents that are similar to the patent document to be evaluated;

[0034] Step 102: using each claim item of the patent document to be evaluated to traverse each similar patent document to perform technical feature comparison, and obtain the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document;

[0035] Step 103: Draw a technical feature spectrum diagram according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum diagram corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum diagram corresponds to the similarity of technical features between each claim item and each similar patent document, and the size of the similarity is corresponding to a preset color tone;

[0036] Step 104: Preliminary evaluation is performed on the patent document to be evaluated according to the technical feature spectrum to obtain a preliminary evaluation result of the patent document to be evaluated.

[0037] Compared with the prior art technical solution of pre-evaluating the patent documents to be evaluated by manually comparing the similarity of the documents, which is inefficient and less intuitive, the patent document pre-evaluation method provided by the embodiment of the present invention has the following beneficial technical effects:

[0038] First, the embodiment of the present invention inputs the information of the patent document to be evaluated into a pre-established semantic analysis model, retrieves multiple similar patent documents similar to the patent document to be evaluated, realizes efficient retrieval through semantic analysis, and lays a solid foundation for the subsequent efficient preliminary evaluation of the patent document.

[0039] Secondly, the embodiment of the present invention uses a traversal algorithm to traverse each claim of the patent document to be evaluated through each similar patent document for technical feature comparison, and obtains the similarity of technical features of each claim of the patent document to be evaluated and each similar patent document, thereby realizing the use of the traversal algorithm for technical feature comparison: the patent document to be evaluated traverses all claims, and the similar patent document traverses all document contents. This algorithm logic is close to the business logic used by users, so that patent documents can be accurately pre-evaluated.

[0040] Again, in an embodiment of the present invention, a technical feature spectrum is drawn according to the similarity of technical features between each claim of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim and each similar patent document, and the size of the similarity is corresponding to a preset color tone; based on the technical feature spectrum, the patent document to be evaluated is pre-evaluated to obtain a pre-evaluation result of the patent document to be evaluated, thereby realizing a technical feature spectrum drawn based on the technical feature similarity, and performing visual interaction, so that the pre-evaluation result of the patent document to be evaluated can be intuitively obtained.

[0041] In summary, the patent document pre-evaluation method provided in the embodiment of the present invention can pre-evaluate the patent document efficiently, accurately and intuitively. Figures 2 to 6 The preliminary evaluation method of the patent document is introduced in detail.

[0042] 1. First, the above step 101 is introduced, that is, the step of performing intelligent retrieval based on semantic analysis.

[0043] First, the construction of the semantic web is introduced, that is, the semantic analysis model is established in advance.

[0044] 1. Model selection

[0045] The construction of the semantic web uses training data to build semantic relationships between semantic elements. Its function is to realize the vector representation of vocabulary to facilitate subsequent semantic calculations.

[0046] At present, the training of semantic web mostly uses semantic models based on neural networks. The embodiment of the present invention uses FASTTEXT and BERT as training models of semantic web. That is, in one embodiment, the semantic analysis model includes FASTTEXT and BERT models, and the WORD2VECT model (classic model) is investigated.

[0047] The characteristics of the three models in the embodiment of the present invention are shown in Table 1 below:

[0048]

[0049]

[0050] Table 1

[0051] The embodiment of the present invention does not use the word2vect model, but selects fasttext and bert as the semantic web construction model. The reason is:

[0052] 1) Although word2vect is a classic semantic model, it is relatively old, and its basic theoretical basis is a linear statistical model rather than a neural network. Its adjustment and improvement space is very limited. The inventor has also used the word2vect model in other studies, and its vectorization results are slightly insufficient for semantic comparison.

[0053] 2) The model that the embodiment of the present invention is most inclined to try to use is the BERT model. There are three reasons: First, the BERT model has the largest neural network scale and the largest word vector scale, which means that for each word, the vector expression granularity in the BERT model is the finest. The model has the most possibilities for adjustment. Second, there are many social studies on the BERT model, and there are a large number of pre-training results that can be directly applied, which means that the research starting point of the embodiment of the present invention does not have to start from scratch. The results of the embodiment of the present invention can also be provided to subsequent researchers. Third, the European Patent Office used the Google-based BERT pre-training results in its automatic classification project and conducted retraining. Its application effect in the field of English patents is good.

[0054] 3) The reason why the embodiment of the present invention still chooses to use the FASTTEXT model first. Because the fasttext model is between word2vect and bert in terms of parameters such as training time, neural network model, and vector scale. Its training time is actually closer to word2vect, but its performance is close to bert. It is actually a very efficient neural network model. The long training time of bert is a very important influencing factor. Since the embodiment of the present invention is based on the construction of the semantic web, it is necessary to carry out subsequent technical feature selection, semantic comparison, visualization experiments, empirical research and other related work in sequence. Therefore, in the semantic web link, there needs to be a relatively reliable model that can quickly obtain results as a guarantee.

[0055] In the specific implementation example, the correctness of the inventor's choice of FastText was also proved in the actual research process. Because the first BERT training used a BERT model pre-trained based on network resources, its training effect was not as good as that of FastText in terms of the comparison effect of calculating patent sentences. However, this BERT training consumed nearly 2 months of the inventor's time.

[0056] When studying the specific embodiments, a very important key problem was found in the process of studying the embodiments of the present invention: semantic comparison in patent examination often requires comparison and calculation of the "closeness" and "similarity" of two technical description texts. At this stage, the basic theoretical basis of semantic network training based on neural networks is based on the text relationship statistics of patent texts. For example, if two words appear in the same sentence, they are judged to have semantic associations and are included in the statistics (neural networks are also statistics). The statistical relationship is actually a "context association" relationship. This is different from the "similarity" we need in patent examination. Admittedly, when two words, if their context associations are highly similar from a statistical perspective, the word vectors formed will be very close, and the calculation of "similarity" can be achieved to a certain extent. But after all, the business scenario requirements and the underlying logical basis of technology are different. If you want to truly achieve "specialization" in the comparison of "technical similarity", you need to specially select the training corpus of the underlying model, which has been tried within the scope of the capabilities of the embodiments of the present invention.

[0057] 2. Word segmentation strategy

[0058] Word segmentation is a necessary link in all Chinese semantic systems. Its function is to use a dictionary to divide Chinese sentences into vocabulary units and mark the part of speech of each vocabulary unit.

[0059] The word segmentation strategy currently used in the embodiment of the present invention is as follows:

[0060] 1) According to the preset dictionary classification strategy.

[0061] Currently, FASTTEXT uses a preset technical dictionary for classification. The preset technical dictionary is a technical dictionary obtained through years of summary based on patent data and extracted using a semantic engine (including new word discovery function).

[0062] 2) No word segmentation (character segmentation) strategy.

[0063] At present, the BERT model adopts a word-free segmentation strategy, that is, a word-by-word segmentation strategy. This strategy also does not mark the part of speech. The basic concept of this strategy is that the smallest unit of Chinese is a word, and any form of word segmentation may lose or mislead semantics. Therefore, a word-by-word segmentation strategy is used to train the semantic model using ordered combinations of words. The amount of training calculation is geometrically multiple of that of normal word segmentation. Only the BERT model has performed relatively well for this type of word segmentation strategy (other models have made similar attempts, but the results are not good). Therefore, in the BERT training of the embodiments of the present invention, a word-free segmentation strategy is used.

[0064] 3) Redundant word segmentation strategy (not adopted after the experiment).

[0065] The redundant word segmentation strategy is to perform redundant word segmentation on a long compound word. For example, after word segmentation, "electric car" will become "electric car" "," electric "," car "." The advantage of redundant word segmentation is that it can avoid semantic ambiguity caused by word segmentation errors to a certain extent, and intelligent retrieval using redundant word segmentation will have a good effect. The disadvantage is that it will increase the amount of calculation by several times to dozens of times. In the FASTTEXT model training of the embodiment of the present invention, the strategy of redundant word segmentation was tried. However, since there was almost no overall improvement in the effect, the strategy was eventually abandoned. Note: The reason why the redundant word segmentation effect is not obvious is that we use the full-text patent text segmentation training. Although the effect of redundant word segmentation for a single long word is very obvious, the redundant words often exist in the full-text word segmentation, and their contexts are similar, so the cost performance is extremely low.

[0066] 4) Short word strategy.

[0067] The FSATTEXT model of the embodiment of the present invention uses a short word segmentation strategy. The reason is that the patent field is a field where many new words appear, with the number of new words being about one million each year. Too many new words will lead to frequent and unstable updates of the segmentation dictionary. However, new words in the patent field are often in the form of compound words, which are new words composed of multiple technical terms. For example, "electric car" is segmented into "electric", "car", or "electric car", and the loss itself is not large at the semantic level. Therefore, in the FASTTEXT model of the embodiment of the present invention, in order to keep the dictionary in a relatively stable state, a short word segmentation strategy (words with less than five characters) is used.

[0068] 5) Noun strategy

[0069] In the FASTTEXT model of the current embodiment of the present invention, the strategy used in feature selection is to select noun parts of speech after word segmentation. This is because nouns contain the vast majority of technical vocabulary. The vocabulary of other parts of speech has a lower relevance to technology. In fact, part of the reason why noun parts of speech are selected in the embodiment of the present invention is that the workload and experimental time are limited, and no more attempts are made. In technical exchanges, the inventor believes that if verbs are also included in the subsequent feature selection, the effect may be improved.

[0070] 3. Training data selection

[0071] One of the training sets selected in the embodiment of the present invention may be invention patent data and utility model patent data publicly announced in the past five years. This data set is used for training the FASTTEXT model and the first layer training of the BERT model.

[0072] The data range of the past five years was selected because: first, the invention patent data and utility model patent data of the past five years totaled more than 18 million, which is rich enough and covers a complete range of technical fields, and the richness of the corpus has fully met the needs of model training. Secondly, as language expression methods and technical terms will change over time, choosing more recent data as training corpus will help the trained model to be more suitable for the current language environment.

[0073] The second training set selected in the embodiment of the present invention can be the corresponding chapter fragments of the comparison documents (comparative documents) marked by the examiner in the search report. This training set is used for the second-level training of the BERT model, which can further improve the accuracy of subsequent intelligent retrieval.

[0074] The reason for choosing this training data set is that conventional patent data, as corpus, trains "semantic relevance" relationships rather than "technical similarity" relationships. Therefore, the underlying logic of the semantic network trained using conventional corpus does not fully conform to the judgement logic of the examiner when comparing documents. Therefore, it is necessary to discover corpus with completely consistent underlying logic for training.

[0075] This corpus is only used as a supplementary training corpus for the BERT model because its basic quantity is insufficient and it is difficult to build a complete semantic model.

[0076] Another problem with this type of corpus is that it is extremely difficult to collect and organize. Although it is mentioned in the search report, what is actually mentioned in the search report is the directional information of the corpus pair, not the actual corpus. The directional information needs to be extracted, and then the corresponding corpus needs to be located based on the directional information and then extracted. Moreover, this directional information is mixed in unstructured data such as tables and texts in search reports or notices. How to completely and massively extract these corpora for training has become a new problem.

[0077] In the embodiment of the present invention, a relatively simplified corpus extraction rule is temporarily used. It is as follows:

[0078] 1) Select a patent with only one independent claim and X and Y comparative literature (the inventor has tried to select the above patent with only one claim, but the amount of data is too small to meet the training needs).

[0079] 2) Extract the corresponding claims from the search report and compare the document location information (document number, paragraph position). Exclude patent data without the above precise location information.

[0080] 3) Through the above-mentioned pointing information, go to the two patents, extract the text paragraphs at the corresponding positions respectively, and piece them into corpus pairs.

[0081] According to the above rules, the embodiment of the present invention has currently extracted more than 8,000 pairs of patent texts as training corpus.

[0082] 4. Application of Semantic Model Structure

[0083] In addition to the above-mentioned customized model learning for business, the embodiment of the present invention also proposes a sub-structure application model of the BERT model.

[0084] The BERT model has a characteristic that when the training corpus is different, the construction results of its semantic network are completely different. On the basis of one training, a second training can be performed. The first BERT model training of the embodiment of the present invention is carried out on the basis of the existing open source shared BERT model. The reason for this choice is that in the first stage, the inventor believes that the patent, as the only training corpus, has a semantic system that is too biased towards the patent-specific description method, that is, the semantic network is incomplete. If only patent data is used for training, it is likely to cause the semantic network to be too one-sided. For example, synonyms in patents rarely appear in the same document. This results in extremely low correlation between synonyms after semantic network training. Therefore, the first BERT model training of the embodiment of the present invention uses an existing model that has been trained, and patent data training is carried out on this basis.

[0085] However, the results of the first training were very poor. The reason for this is that the original model was trained on news and media corpora with a volume of hundreds of millions, which is much larger than the patent corpus. Therefore, the role of the patent corpus was very small.

[0086] But inspired by this, the inventor proposed a sub-structured application model of the BERT model. The specific idea is: subdivide the patent corpus according to its different focuses, such as title (focusing on the theme of the invention), abstract (focusing on the key points of the invention), claims (focusing on complete technical features and protection points), background technology (focusing on fields and effects), utility sentences (focusing on effects), and embodiments (focusing on implementation cases). And discard the illustrations (the semantic structure is too special). Train different bert models for different parts. When using it, since the usage scenarios are often also for patents, it is also possible to first disassemble the parts according to the user object, and then describe the semantic model. This objectively forms an effect of distinguishing the use of semantic features.

[0087] Secondly, the selection of technical feature units is introduced, that is, the feature units selected when comparing the similarities between two documents during retrieval.

[0088] In the embodiment of the present invention, the selection of the technical feature unit mainly refers to the minimum technical feature description unit used for comparison.

[0089] The selection of the minimum technical feature description unit generally has the following levels:

[0090] 1. Vocabulary level. That is, using technical words as the technical feature description unit of the patent. For example, patent CN111273180A "A lithium battery lithium deposition detection method and device" uses vocabulary-level feature units, which is to compare words such as "lithium battery", "device", "lithium deposition", "charging", etc.

[0091] 2. Phrase level. That is, phrases are used as the technical feature description unit of the patent. For example, patent CN111273180A "A lithium battery lithium deposition detection method and device" uses phrase-level feature units, which is to compare phrases such as "lithium deposition detection method and device", "improving the reliability of lithium deposition diagnosis", "induced pressure information", etc.

[0092] 3. Short sentence level. That is, short sentences are used as the technical feature description unit of the patent. For example, patent CN111273180A "A lithium battery lithium deposition detection method and device" uses short sentence level feature units, which is to compare short sentences such as "This application provides a lithium battery lithium deposition detection method and device", "involving the field of electric vehicle technology", "used to improve the reliability of lithium deposition diagnosis", etc.

[0093] 4. Paragraph level. That is, use paragraphs as the technical feature description unit of the patent. For example, patent CN111273180A "A lithium battery lithium plating detection method and device" uses paragraph-level feature units, which is to compare "The present application provides a lithium battery lithium plating detection method and device, which relates to the field of electric vehicle technology and is used to improve the reliability of lithium plating diagnosis. The method includes: using at least two charging rates to cyclically charge the lithium battery, the at least two charging rates include a first charging rate and a second charging rate, the first charging rate is greater than the second charging rate and the difference between the first charging rate and the second charging rate is greater than a preset rate threshold; in the cyclic charging process, respectively detect the state parameters of the lithium battery, and obtain at least two corresponding pressure information; according to the at least two pressure information, determine the pressure information caused by the lithium plating effect of the lithium battery, the pressure information caused by the lithium plating effect is used to indicate the pressure value change caused by the lithium plating effect of the lithium battery at different SOCs; when there are multiple continuous pressure values ​​corresponding to the SOC in the pressure information caused by the lithium plating effect that are greater than zero, it is determined that the lithium battery has lithium plating." Such paragraphs.

[0094] 5. Others. In addition, special rules such as paradigm structures and long sentences can also be used to agree on the technical feature description units for comparison.

[0095] The technical feature unit selection strategy finally adopted by the embodiment of the present invention. This strategy is a special rule: select the claim item as the technical feature unit. The claim item is used as the technical feature unit, and its technical feature unit splitting rule is the splitting rule of the continuation of the long sentence, with the period and other long sentence terminators as the segmentation markers. However, its expression is at the paragraph level, because its text volume is obviously different from the long sentence text volume of the main text. In addition, the strategy of using the claim item as the technical feature unit also adds a special semantic processing to eliminate the non-technical expression link, that is, the text content before "characterized by".

[0096] In addition, the selection of the above-mentioned technical feature units is a rule for one side of the comparison, that is, the patent document to be examined (the patent document to be evaluated is the patent document to be examined in this case). For the patent document used for comparison, the selection strategy of its technical feature units is different. The text part of its claims still uses the strategy of using the claims as technical feature units. However, the full text part uses the "long sentence strategy + scale equivalence strategy". The long sentence strategy will not be elaborated. The scale equivalence strategy refers to the length of the text used for comparison, which is composed of multiple long sentences. The length of the spliced ​​text is generally lower than the paragraph length of the main text, but is equivalent to the text length of the claims.

[0097] In specific implementation, the reason why word-level and phrase-level technical feature units are not used is that when a patent describes a technical solution, simple words and phrases will lose a lot of semantic information.

[0098] In specific implementation, the reason why short-sentence-level technical feature units are not used is that patents often contain a large number of parallel short sentences, whose semantic information content is not much different from that of words and phrases.

[0099] In specific implementation, the reason for choosing the entire sentence of the claim as the technical feature unit is that it can retain the complete semantic content of a single innovative feature in the technical solution to the greatest extent. At the same time, the distinction between different innovative features is also the clearest.

[0100] In specific implementation, the scale equivalence strategy is used in the selection of technical feature units in the comparative literature. This is because the description granularity of long sentences and paragraphs in the claims and the text is not equal. The scale of long sentences in the claims is often much larger than that of long sentences in the text, but much smaller than that of paragraphs in the text. The use of long sentences or paragraph strategies in the text may cause the semantic scale of the two sides of the semantic comparison to be unequal, and lose the meaning of comparison.

[0101] The comparison between technical feature units is introduced again.

[0102] The comparison between technical feature units is to vectorize the technical units and then use the cosine similarity formula for calculation.

[0103] Cosine similarity measures the similarity between two vectors by measuring the cosine of the angle between them. The cosine of a 0-degree angle is 1, while the cosine of any other angle is no greater than 1; and its minimum value is -1. Therefore, the cosine of the angle between two vectors determines whether the two vectors point in roughly the same direction. When two vectors have the same direction, the cosine similarity value is 1; when the angle between the two vectors is 90°, the cosine similarity value is 0; when the two vectors point in completely opposite directions, the cosine similarity value is -1. This result is independent of the length of the vector, but only related to the direction in which the vector points. Cosine similarity is usually used in positive space, so the value given is between -1 and 1.

[0104] Note that these bounds apply to vector spaces of any dimension, and that cosine similarity is most commonly used in high-dimensional positive spaces. For example, in information retrieval, each term is assigned a different dimension, and a dimension is represented by a vector whose values ​​on each dimension correspond to the frequency of the term in the document. Cosine similarity can therefore give the similarity of two documents in terms of their topics. It is often used for document comparison in text mining.

[0105] Given two attribute vectors, A and B, their cosine similarity θ is given by the dot product and the length of the vectors as shown in the following formula.

[0106]

[0107] Here i represents the components of vectors A and B respectively.

[0108] In one embodiment, inputting the information of the patent document to be evaluated into a pre-established semantic analysis model to retrieve multiple similar patent documents similar to the patent document to be evaluated may include: retrieving each similar patent document similar to the patent document to be evaluated according to the following method:

[0109] Divide the patent document to be evaluated and each document to be compared into multiple parts;

[0110] Performing semantic analysis on each part to obtain the semantic analysis result of each part;

[0111] Determine the weight value of each part of the patent document to be evaluated and each part of each document to be compared according to the semantic analysis results of each part;

[0112] According to the weight value of each part of the patent document to be evaluated and each document to be compared, a weighted average result of the patent document to be evaluated and each document to be compared is obtained;

[0113] Determine the similarity between the patent document to be evaluated and the document to be compared based on the weighted average results of the patent document to be evaluated and each document to be compared;

[0114] When the similarity between the patent document to be evaluated and the document to be compared is greater than a preset value, the document to be compared is regarded as an approximate patent document of the patent document to be evaluated.

[0115] In specific implementation, the above detailed implementation scheme of obtaining each similar patent document similar to the patent document to be evaluated based on semantic analysis further improves the efficiency of patent document evaluation.

[0116] In specific implementation, the above-mentioned "part" may be an abstract, claims, description, etc.

[0117] In specific implementation, the patent document to be evaluated can be the patent application document before the patent application is submitted, or it can be the patent document to be reviewed, or it can be the patent document that has been authorized. The application scenarios of the embodiments of the present invention may include: self-preliminary review by colleges and universities, auxiliary writing (cultivation), user-friendly review assistance for examiners (opening up the functions of custom adding comparison documents and automatic matching comparison documents + novelty visualization), and many other scenarios. Two examples of the application scenarios of the embodiments of the present invention are given as follows.

[0118] Scenario 1: Rating the patentability of patent documents: In this step, users submit technical documents (patent document information to be evaluated) which can be 1) names, 2) claims, or key technical features of technical documents, one paragraph is one feature, and 3) the full text of technical documents. With the above three simple inputs, you can submit for preliminary evaluation.

[0119] Scenario 2: Invalidation analysis of patented documents: In this step, users submit the 1) name, 2) claim item, 3) full text of the technical document (patent document information to be evaluated), and 4) patent application time. Or users can directly enter the patent number for quick analysis. The above input can be submitted for invalidation analysis.

[0120] In specific implementation, step 101 is a feature comparison algorithm based on semantic analysis technology. The algorithm constructs a domain patent knowledge base, calculates the patent semantic weight of the term, represents the patent text to be compared as a vector with patent semantic information, and further represents the patent text as a patent text vector with position weight information in combination with the position of the term in the patent document structure, and finally represents the patent text as a patent text vector with patent semantic weight information, position weight information and word frequency weight information. Through the semantic comparison of the vector set of the text to be compared and the vector set of the patented text, the feasibility of this algorithm in the intelligent judgment of novelty is studied.

[0121] The research on feature comparison algorithm based on semantic analysis technology is one of the core research points of the embodiment of the present invention. This research point is divided into three parts:

[0122] First, we use the most advanced semantic and self-learning technologies to build a semantic space using patent data. The semantic space has the ability to locate and connect words.

[0123] Second, segment the technical features of the patent, and strive to clearly and completely express the technical features of the patent in the semantic computing model.

[0124] Third, make an overall comparison of the features. The comparison should be as close to the "technical essence" as possible, rather than a literal comparison.

[0125] Therefore, from the above, we can see that when performing intelligent retrieval, in addition to considering semantic weight, we can also consider word frequency weight, position weight, etc. to further improve the accuracy of retrieval.

[0126] Connection with patent intelligent retrieval: Conduct a survey of existing intelligent retrieval methods based on patent databases that are suitable for patent review business scenarios, as well as the application of existing similarity retrieval technologies, to determine appropriate intelligent retrieval technologies as the basis for subsequent semantic comparison application research. This research content aims to separate the application scenario of "patent technical solution comparison within a specific data range" from the application scenario of "intelligent retrieval of massive patent databases" and standardize the specific technical parameter requirements between the two. This allows the research results of the embodiments of the present invention to be more flexibly connected to the existing intelligent retrieval system.

[0127] 2. Next, the above step 102 is introduced, that is, comparing and traversing the features of the patent document to be evaluated (such as the document to be reviewed) and the intelligent search result set (a set of multiple similar patent documents).

[0128] like Figure 2 As shown, the traversal steps may include:

[0129] Step 1: Obtain a comparison patent collection from the outside (in the experiment of the embodiment of the present invention, the first 50 patents are obtained from the intelligent retrieval system engine).

[0130] Step 2: Decompose the claim items of the pending document into features (In the embodiment of the present invention, the claim items are decomposed into sentences, as described in the following embodiment for details.

[0131] Step 3: For all text contents of the specifications of all patents in the patent collection (including claims and full text of the specifications), feature decomposition is performed using a granularity approach similar to that of the claims. For details, see the introduction of the following embodiment.

[0132] Step 4: Select the decomposed features in order according to the order of the claims in the document to be examined.

[0133] Step 5: Select the comparison patents in the order of the comparison patent set.

[0134] Step 6: Select the decomposed features in order according to the order of the compared patents.

[0135] Step 7: Use the claim features of the pending document selected in step 4 and the comparison patent features selected in step 6 to perform feature comparison.

[0136] Step 8: Determine whether all features of the compared patent have been traversed. If not, return to step 6. If yes, proceed to step 9.

[0137] Step 9: Use the maximum value of all comparison results as the final comparison value of the claim feature.

[0138] Step 10: Determine whether all the compared patents have been traversed, if not, return to step 5. If yes, proceed to step 11.

[0139] Step 11: Determine whether all features of the document to be reviewed have been compared. If not, return to step 4. If so, proceed to step 12.

[0140] Step 12: The comparison traversal is completed.

[0141] In specific implementation, because only a complete claim in the patent claim can truly and completely express a minimum unit of technical features, rather than technical feature fragments, the embodiment of the present invention finally adopts the sentence-by-sentence decomposition of the claim as the minimum feature unit for initiating comparison.

[0142] In one embodiment, using each claim item of the patent document to be evaluated to traverse each similar patent document to perform technical feature comparison, and obtaining the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document, may include: using each claim item of the patent document to be evaluated to traverse each similar patent document to perform full-text technical feature comparison, and obtaining the similarity of the technical features with the greatest similarity between each claim item in the patent document to be evaluated and each similar patent document;

[0143] Drawing a technical feature spectrum diagram based on the similarity of technical features between each claim item in the patent document to be evaluated and each similar patent document may include: drawing a technical feature spectrum diagram based on the similarity of technical features between each claim item in the patent document to be evaluated and each similar patent document with the greatest similarity in technical features; the intersection of each claim item and each similar patent document in the technical feature spectrum diagram corresponds to the similarity of technical features between each claim item and each similar patent document with the greatest similarity in technical features.

[0144] In specific implementation, all claims are compared with all documents of similar patent documents (comparative documents, comparative documents), and the highest value (the technical feature with the greatest similarity between each claim in the patent document to be evaluated and the technical feature in each similar patent document) is used as the comparison result. This is convenient for directly locating the feature comparison point. Of course, in specific implementation, a comprehensive feature vector can also be calculated for a similar patent document, and the feature vector of a claim can be compared with the comprehensive feature vector to obtain the similarity of the technical features of the two.

[0145] In one embodiment, each claim item of the patent document to be evaluated is used to traverse each similar patent document to compare technical features, and the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document is obtained, which may include:

[0146] According to the size of each claim item of the patent document to be evaluated, the size of the comparison feature of each similar patent document is adjusted to obtain an adjustment result;

[0147] According to the adjustment result, each claim item of the patent document to be evaluated is used to traverse each similar patent document for technical feature comparison, so as to obtain the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document.

[0148] In specific implementation, the full text of a patent is different from the patent claims, and its sentence expressions can be very long or very short, and there is no particularly clear method to distinguish the completeness and granularity of its technical feature expressions. In semantic comparison, this will lead to a practical operational problem, that is, the sentence features decomposed from the claims are mostly longer, while the sentence features decomposed from the full text are mostly shorter. If the two are compared directly, there is actually a problem of unequal granularity. Therefore, the method adopted in the embodiment of the present invention is to use the sentence as the smallest unit, and use the feature scale of the claim to adjust the size of the features in the approximate patent document, to ensure that the feature scale (the scale can refer to the number of words, etc.) of the two sides of the comparison is basically equal, which can improve the accuracy of the comparison, and thus improve the accuracy of the pre-evaluation of the patent document.

[0149] In summary, the advantages of this traversal algorithm are:

[0150] 1. All claims of pending patents are traversed, and all document contents of comparative documents are traversed. The logic of this algorithm is similar to the business logic used by users, such as examiners, during review.

[0151] 2. All claims are compared with all documents of the referenced patent, and the highest value is used as the comparison result, which is convenient for directly locating feature comparison points.

[0152] 3. Again, for ease of understanding, the above steps 103 and 104 are introduced together below, which are the steps of performing visual interactive evaluation based on the drawn technical feature spectrum, that is, the visual presentation algorithm and interactive design of the comparison results.

[0153] In one embodiment, the patent document to be evaluated is pre-evaluated according to the technical feature spectrum to obtain the pre-evaluation result of the patent document to be evaluated, which may include:

[0154] When receiving a click operation of the user on the intersection, a preliminary evaluation result of comparing a single claim of the patent document to be evaluated with a single similar patent document is obtained;

[0155] Or, when receiving a user's click operation on a preset ordinate on the ordinate axis, a preliminary evaluation result of comparing a single claim of the patent document to be evaluated with all similar patent documents is obtained;

[0156] Alternatively, when a user clicks on a preset horizontal axis on the horizontal axis, a preliminary evaluation result of comparing all claims of the patent document to be evaluated with a single similar patent document is obtained.

[0157] During specific implementation, click on a specific square (intersection, which can be in the form of a square or a circle, etc.) to compare a single feature (claim item) with a single patent, click on the vertical axis to compare a single feature (claim item) with all similar patents, and click on the horizontal axis to compare all features (claim items) with a single patent.

[0158] In one embodiment, pre-evaluating the patent document to be evaluated according to the technical feature spectrum to obtain the pre-evaluation result of the patent document to be evaluated may also include:

[0159] When a user clicks on an extreme value of feature similarity in the technical feature spectrum graph, a preliminary evaluation result is obtained by comparing all claims of the patent document to be evaluated with an approximate patent document having the greatest similarity to each claim.

[0160] During specific implementation, click on the legend “Feature Similarity Extreme Value” to compare all features (claim items) with their respective corresponding patents with the greatest similarity.

[0161] In specific implementation, the above-mentioned interactive design based on graphics can provide users with more detailed basis for novelty judgment and patent data of different granularity, so that users can have more accurate control over the positioning and differences of the technical document in the current patent library. Users can obtain more targeted inspiration about the technology to be applied for patents from the comparison chart with similar patents, which is convenient for accurate modification of technical solutions and adjustment of patent text language; it is convenient for users to judge the novelty of the claims of the patents to be applied for and explore the patentability prospects of the patents to be applied for from fine-grained data.

[0162] During specific implementation, the indicators involved in the embodiments of the present invention are described as follows.

[0163] 1. How to obtain the top 50 articles: Based on the claims and the description, you can combine them into a whole to perform a similarity search of the entire library, and then sort the top 50 articles in reverse order of similarity scores.

[0164] 2. What is "similarity"? The "similarity" in the current table is the ranking score given by the system when searching for similarity of the entire text. Later, we will map the similarity score to 0-100 points.

[0165] 3. Explanation of other indicators: By comparing each claim with the 50 patents obtained sentence by sentence, a "technology spectrum chart" is drawn.

[0166] 4. Comprehensive novelty of the entire document: Analyze and evaluate the comprehensive patentability of the application document from the perspective of the entire technical spectrum.

[0167] 5. Maximum novelty of feature: The one with the highest novelty among multiple claims (features), so that the applicant can adjust and strategically reorganize the claims based on the data and spectrum.

[0168] 6. Compare patent damping coefficients: Find the most likely patent to conflict with the application based on feature comparison data, and list the comparison data. For example, if a preliminary assessment is conducted on the applied patent, the damping coefficient is likely to be 1. The closer it is to 1, the more likely it is to conflict.

[0169] 7. Depth of competition: the degree of conflict between the 50 most similar patents and this application.

[0170] 8. Breadth of competition: The degree of conflict between all data retrieved during similarity search and the application document based on the full text of the application document.

[0171] 9. Technical coupling: the possible degree of connection between this application and other patents.

[0172] like Figure 3As shown in the figure, similar documents of a patent document are displayed in the form of a spectrum. The higher the similarity, the greater the thermal value of the corresponding point on the spectrum (the heavier the red). If the similarity is lower, the thermal value corresponding to the point on the spectrum is smaller and the color is lighter, such as light blue. Since there can be no color in patent drawings, Figure 3 Colors are not shown, but those skilled in the art can imagine that a table of different similarity sizes, thermal values, and tones can be used to draw a technical feature spectrum. In addition, in addition to using different tones to represent different similarity sizes, other forms can also be used, such as different graphics, etc.

[0173] The technical features of the target technical document are calculated through semantic comparison of the embodiment of the present invention, and the results are displayed flatly on one screen in a two-dimensional manner. The vertical axis corresponds to the claims of the patent application document, and the horizontal axis corresponds to N = 50 most similar patents. The intersection of the horizontal and vertical axes is the claim. For the corresponding patent document, the feature comparison calculation is traversed sentence by sentence and paragraph by paragraph for the entire text to obtain the highest comparison similarity value. Use a gradient of cold and warm tones to express the similarity of features from low to high, thereby achieving a macroscopic observation and positioning effect.

[0174] From the overall color warmth and coldness, we can get a macro view of the overall technical novelty of the application document. Using the vertical color as a clue, we can find the comparison documents or conflicting applications that may cause technical innovation conflicts; using the horizontal color as a clue, we can find the "high-quality" innovation points of the application document. We can even string together relatively scattered comparison points to find the possibility of combined conflicts.

[0175] In specific implementation, the spectrum can be used to "see through the novelty of technical features with three eyes". The first look is to intuitively see the overall technical novelty of the document to be applied for from the overall warm and cold color conditions; the second look is to use the vertical color as a clue to find the comparison documents or conflicting applications that may cause technical innovation conflicts; the third look is to use the horizontal color as a clue to find the "high-quality" innovation points of the document to be applied for.

[0176] In specific implementation, the biggest advantage of the embodiment of the present invention is the high compression of information, which brings about the improvement of macro judgment ability. Currently, there is no similar product that can provide clear clues for novelty judgment based on the simultaneous browsing of 50 patents.

[0177] In specific implementation, the embodiment of the present invention checks the overall fit of each claim item in the technical document with similar patents, analyzes the technical features of the patent application text and the 50 patents most similar to it, and compares the features. And highly compresses the relevant information in a picture in the form of color. Use warm and cold colors and feature arrangement to show the novelty of the overall technical features of the patent application.

[0178] In specific implementation, step 103 and step 104 are based on the research of feature comparison application scenarios of visualization interaction technology: visualization interaction application research, research on how to use visualization technology to provide clear clues for novelty judgment based on the simultaneous browsing of dozens of patents. Research on how to highly compress a large amount of comparison information to improve the examiner's macro judgment ability in the patent examination process.

[0179] At present, the research on artificial intelligence technology related to patent retrieval still follows the original manual work application mode, and there are few research cases that break through the existing application mode. The embodiment of the present invention intends to conduct innovative research on the comprehensive comparison and browsing of a small number of patent collections, and try to study the changes in business forms brought about by visualization technology.

[0180] 4. Finally, further optimization steps are introduced.

[0181] In specific implementation, the embodiment of the present invention relies on rich intellectual property data and the accumulation of strong and reliable patent evaluation capabilities. The patent application pre-evaluation system uses patent big data through machine learning to automatically understand the semantics of technical documents that have not yet been patented, and searches for patents similar to its technical solutions in the entire database, thereby obtaining a similar patent database of the technology, and realizing the evaluation of the technical document, such as Figure 4 and Figure 5 As shown, the evaluation results can include three parts: novelty, competitive situation, and business prospects.

[0182] In one embodiment, the patent document pre-evaluation method may further include: pre-evaluating the competition situation and business prospects of the patent document to be evaluated, and obtaining a pre-evaluation result of the competition situation and business prospects of the patent document to be evaluated;

[0183] Preliminarily evaluating the patent document to be evaluated according to the technical feature spectrum to obtain a preliminarily evaluating result of the patent document to be evaluated may include: preliminarily evaluating the novelty of the patent document to be evaluated according to the technical feature spectrum to obtain a preliminarily evaluating result of the novelty of the patent document to be evaluated;

[0184] The patent document pre-evaluation method may further include: obtaining a comprehensive pre-evaluation result of the patent document to be evaluated based on the novelty pre-evaluation result of the patent document to be evaluated, and the competition situation and business prospect pre-evaluation results.

[0185] When implementing it, Figure 4 and Figure 5As shown, in an embodiment of the present invention, in addition to pre-evaluating the novelty of the patent document to be evaluated according to the technical feature spectrum diagram to obtain the novelty pre-evaluation result of the patent document to be evaluated, the competitive situation and business prospects of the patent document to be evaluated can also be pre-evaluated to obtain the competitive situation and business prospects pre-evaluation result of the patent document to be evaluated; the pre-evaluation result of the novelty of the patent document to be evaluated, as well as the pre-evaluation result of the competitive situation and business prospects can be combined to obtain a comprehensive pre-evaluation result of the patent document to be evaluated, for example, the comprehensive pre-evaluation result can be A-level, B-level, etc., which is convenient for users to fully understand the patentability and various aspects of their own patent documents, and facilitate the smooth progress of subsequent patent applications or invalidation procedures.

[0186] In specific implementation, after submitting the technical document (patent) to be evaluated, after a certain period of intelligent evaluation, the preliminary evaluation of the patent application (technical document) can not only evaluate the novelty, competitive situation, and business prospects of the technical document, but also obtain a database of similar technical patents. Each similar patent has separate evaluation data, and the evaluation process data of each patent can be viewed; there is also statistical evaluation data of the entire similar patent library, such as IPC classification, which can predict the possible classification of the technical text and even obtain information such as the technical prospects and business prospects of the classification, ultimately giving users clear patent knowledge feedback.

[0187] In specific implementation, the embodiment of the present invention will conduct semantic understanding, multi-dimensional feature analysis, and comprehensive data calculation for each technical document, and finally give a comprehensive rating to the target technical document. Users can get straight to the point and quickly obtain the positioning of the technical document from the patent perspective, which is convenient for subsequent decision-making.

[0188] In specific implementation, the embodiment of the present invention also provides a concise evaluation report of the target technical document, which includes the comprehensive rating of the document, the patentability star rating, the competitive situation of the target technical document in the Chinese patent database, and the commercial prospect prediction of the target patent document. It also provides the technical classification of the target technology in the patent field.

[0189] In specific implementation, the technical features of the target technical document (patent document to be evaluated) in the embodiment of the present invention are refracted through a computing prism onto a "screen cloth" based on a similar patent library to obtain a characteristic spectrum of the target technical document. Just as every substance has its own specific characteristic spectrum, every technical document has its own specific characteristic spectrum. The system performs spectral analysis on the target technical document and obtains information such as the patentability of the document and the competitive situation.

[0190] In specific implementation, the embodiment of the present invention provides users with more detailed evaluation data to facilitate users to have a more comprehensive understanding of the evaluation results of the target technical documents. Including 1) novelty, including the comprehensive novelty of the whole article, the maximum novelty of the features, and the damping coefficient of the compared patents, 2) competitiveness, including competition depth, competition breadth, and technical coupling, 3) business prospects, including estimated value, technology popularity, licensing popularity, pledge popularity, transfer popularity, and customs filing popularity.

[0191] During specific implementation, users can view similar patented technologies through the technical document similarity patent library, and can use them to recreate their own technologies or modify existing technical solutions.

[0192] In specific implementation, users can use visual controls to intuitively understand which categories in the patent field the capabilities and technologies of the five dimensions of the commercial prospects of the target technical document may fall into.

[0193] In one embodiment, the above-mentioned patent document pre-evaluation method may further include: adjusting the patent application strategy of the patent document to be evaluated according to the pre-evaluation result of the patent document to be evaluated.

[0194] During specific implementation, the patent application strategy of the patent document to be evaluated is adjusted according to the preliminary evaluation results of the patent document to be evaluated, so as to facilitate the authorization of subsequent patent applications.

[0195] In specific implementation, users can modify / add comparison patents and modify the default value of novelty. Users can customize the addition and deletion of comparison patents on the horizontal axis to meet more professional and personalized functional requirements. Users can also modify the default value of claim similarity judgment. With the participation of users, the accuracy of the preliminary evaluation of patent documents is further improved.

[0196] In order to facilitate understanding of how the present invention is implemented, the following Figure 6 Let me explain this in detail with an example.

[0197] like Figure 6 As shown, the application strategy of "over-splitting technology" is used in this embodiment, that is, the applicant over-splits a certain replaceable technical link of a complete technology that can be patented, and applies for patents multiple times. For example, in this example, heavy metal elements such as Cd, Zn, Cu, etc. including Cd. This phenomenon is very harmful to colleges and universities:

[0198] On the one hand, it increases the cost of meaningless patent applications.

[0199] On the other hand, the scope of protection of the patent that is finally authorized may be locked in an extremely small range, completely losing the protection of the technology and making the technology objectively become "publicly known and publicly used technology."

[0200] Thirdly, it will also cause universities to misjudge the strength of their own patent technology, because such duplicate patents actually only have one core technology.

[0201] Therefore, the effects of the subsequent adjustment of the application strategy are: saving costs, strengthening protection and correctly evaluating technical capabilities.

[0202] Experimental verification: Based on the results of the above technical research, the inventors built an experimental system. In the experimental system, case experiments and evaluations were conducted. The test was conducted to what extent the current semantic comparison technology can judge the novelty of the patent to be applied for. And the visual technology form and interactive experience effect that assist in improving the efficiency of examination. The empirical demonstration of the practical significance of the research results of the embodiments of the present invention in assisting in improving the efficiency of examination.

[0203] In summary, the patent document pre-evaluation method provided by the embodiment of the present invention achieves:

[0204] 1. The embodiment of the present invention implements a semantic comparison algorithm model based on patent text. The semantic comparison result of this algorithm tends to be a comparison of technical substance rather than a comparison at the language level. This feature is consistent with the basic way of thinking of examiners when judging the novelty of patents.

[0205] 2. The embodiments of the present invention have tried to apply advanced semantic web models such as fasttext and BERT, and deep learning algorithms in the professional field of patents. In terms of many technical application details such as model input, model output, training data set, semantic web substructure application, feature length selection, feature selection, word segmentation strategy, etc., targeted designs and experiments have been carried out according to the different characteristics of patent data and the patent industry. Positive results have also been obtained. The embodiments of the present invention disclose all design ideas, technical routes, technical parameters and other information to facilitate subsequent researchers to continue to carry out related work.

[0206] 3. The embodiment of the present invention designs a visual interaction method, which uses visual compression to achieve an innovative solution to the problem of "the last mile of intelligent retrieval". This visual interaction method innovatively solves the problem of rapid macro browsing of dozens of patents and rapid location of approximate features.

[0207] 4. The embodiments of the present invention have conducted case-based and statistical verification studies on patents that are authorized or invalid. The results of the verification studies show that the semantic comparison algorithm and visual interaction design of the embodiments of the present invention have practical application value for the business scenarios of patent review.

[0208] The present invention also provides a pre-evaluation device of the patent document, as described in the following embodiments. Since the principle of the device to solve the problem is similar to the pre-evaluation method of the patent document, the implementation of the device can refer to the implementation of the pre-evaluation method of the patent document, and the repeated parts will not be repeated.

[0209] Figure 7 Schematic diagram of the structure of the pre-evaluation device of the patent document in the embodiment of the present invention, such as Figure 7 As shown, the device comprises:

[0210] Intelligent retrieval unit 01, used to input the information of the patent document to be evaluated into a pre-established semantic analysis model, and retrieve multiple similar patent documents similar to the patent document to be evaluated;

[0211] A traversal feature comparison unit 02 is used to traverse each similar patent document using each claim item of the patent document to be evaluated to compare the technical features, and obtain the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document;

[0212] The spectrum drawing unit 03 is used to draw a technical feature spectrum according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim item and each similar patent document, and the size of the similarity is corresponding to a preset color tone;

[0213] The visual interactive evaluation unit 04 is used to perform a preliminary evaluation on the patent document to be evaluated according to the technical feature spectrum diagram to obtain a preliminary evaluation result of the patent document to be evaluated.

[0214] In one embodiment, the traversal feature comparison unit is specifically used to: use each claim item of the patent document to be evaluated to traverse each similar patent document to compare the technical features of the entire document, and obtain the similarity of the technical features with the greatest similarity between each claim item in the patent document to be evaluated and the technical features in each similar patent document;

[0215] The spectrum drawing unit is specifically used to draw a technical feature spectrum according to the similarity between each claim in the patent document to be evaluated and the technical feature with the greatest similarity in each similar patent document; the intersection point of each claim and each similar patent document in the technical feature spectrum corresponds to the similarity between each claim and the technical feature with the greatest similarity in each similar patent document.

[0216] In one embodiment, the visual interaction evaluation unit is specifically used for:

[0217] When receiving a click operation of the user on the intersection, a preliminary evaluation result of comparing a single claim of the patent document to be evaluated with a single similar patent document is obtained;

[0218] Or, when receiving a user's click operation on a preset ordinate on the ordinate axis, a preliminary evaluation result of comparing a single claim of the patent document to be evaluated with all similar patent documents is obtained;

[0219] Alternatively, when a user clicks on a preset horizontal axis on the horizontal axis, a preliminary evaluation result of comparing all claims of the patent document to be evaluated with a single similar patent document is obtained.

[0220] In one embodiment, the visual interactive evaluation unit is also used to: when receiving a user's click operation on the feature similarity extreme value in the technical feature spectrum graph, obtain a preliminary evaluation result of comparing all claims of the patent document to be evaluated with the approximate patent document with the greatest similarity corresponding to each claim.

[0221] In one embodiment, the intelligent retrieval unit is specifically used to retrieve each similar patent document similar to the patent document to be evaluated according to the following method:

[0222] Divide the patent document to be evaluated and each document to be compared into multiple parts;

[0223] Performing semantic analysis on each part to obtain the semantic analysis result of each part;

[0224] Determine the weight value of each part of the patent document to be evaluated and each part of each document to be compared according to the semantic analysis results of each part;

[0225] According to the weight value of each part of the patent document to be evaluated and each document to be compared, a weighted average result of the patent document to be evaluated and each document to be compared is obtained;

[0226] Determine the similarity between the patent document to be evaluated and the document to be compared based on the weighted average results of the patent document to be evaluated and each document to be compared;

[0227] When the similarity between the patent document to be evaluated and the document to be compared is greater than a preset value, the document to be compared is regarded as an approximate patent document of the patent document to be evaluated.

[0228] It is necessary to be able to input the documents to be applied for according to the different parts of the patent (such as title, abstract, claims, etc.)

[0229] In one embodiment, the patent document pre-evaluation device may further include: a competition situation and business prospect evaluation unit, which is used to: pre-evaluate the competition situation and business prospect of the patent document to be evaluated, and obtain a pre-evaluation result of the competition situation and business prospect of the patent document to be evaluated;

[0230] The traversal feature comparison unit is specifically used to: pre-evaluate the novelty of the patent document to be evaluated according to the technical feature spectrum diagram, and obtain a novelty pre-evaluation result of the patent document to be evaluated;

[0231] The patent document pre-evaluation device may further include: a comprehensive evaluation unit for obtaining a comprehensive pre-evaluation result of the patent document to be evaluated based on the novelty pre-evaluation result of the patent document to be evaluated, and the pre-evaluation results of the competition situation and business prospects.

[0232] In one embodiment, the patent document pre-evaluation method may further include: a strategy adjustment unit, configured to adjust the patent application strategy of the patent document to be evaluated according to the pre-evaluation result of the patent document to be evaluated.

[0233] In one embodiment, the traversal feature comparison unit is specifically used to:

[0234] According to the size of each claim item of the patent document to be evaluated, the size of the comparison feature of each similar patent document is adjusted to obtain an adjustment result;

[0235] According to the adjustment result, each claim item of the patent document to be evaluated is used to traverse each similar patent document for technical feature comparison, so as to obtain the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document.

[0236] An embodiment of the present invention further provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the pre-evaluation method of the above patent document when executing the computer program.

[0237] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the pre-evaluation method of the above patent document is implemented.

[0238] An embodiment of the present invention further provides a computer program product, which includes a computer program. When the computer program is executed by a processor, the pre-evaluation method of the above patent document is implemented.

[0239] Compared with the prior art technical solution of pre-evaluating the patent documents to be evaluated by manually comparing the similarity of the documents, which is inefficient and less intuitive, the pre-evaluation solution for patent documents provided by the embodiment of the present invention has the following beneficial technical effects:

[0240] First, the embodiment of the present invention inputs the information of the patent document to be evaluated into a pre-established semantic analysis model, retrieves multiple similar patent documents similar to the patent document to be evaluated, realizes efficient retrieval through semantic analysis, and lays a solid foundation for the subsequent efficient preliminary evaluation of the patent document.

[0241] Secondly, the embodiment of the present invention uses a traversal algorithm to traverse each claim of the patent document to be evaluated through each similar patent document for technical feature comparison, and obtains the similarity of technical features of each claim of the patent document to be evaluated and each similar patent document, thereby realizing the use of the traversal algorithm for technical feature comparison: the patent document to be evaluated traverses all claims, and the similar patent document traverses all document contents. This algorithm logic is close to the business logic used by users, so that patent documents can be accurately pre-evaluated.

[0242] Again, in an embodiment of the present invention, a technical feature spectrum is drawn according to the similarity of technical features between each claim of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim and each similar patent document, and the size of the similarity is corresponding to a preset color tone; based on the technical feature spectrum, the patent document to be evaluated is pre-evaluated to obtain a pre-evaluation result of the patent document to be evaluated, thereby realizing a technical feature spectrum drawn based on the technical feature similarity, and performing visual interaction, so that the pre-evaluation result of the patent document to be evaluated can be intuitively obtained.

[0243] In summary, the patent document pre-evaluation solution provided by the embodiment of the present invention can pre-evaluate patent documents efficiently, accurately and intuitively.

[0244] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0245] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0246] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0247] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0248] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for pre-evaluating patent documents, characterized in that: include: Input the information of the patent document to be evaluated into a pre-established semantic analysis model to retrieve multiple similar patent documents similar to the patent document to be evaluated; Using each claim item of the patent document to be evaluated to traverse each similar patent document for technical feature comparison, and obtaining the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document, which includes: adjusting the size of the comparison feature of each similar patent document according to the size of each claim item of the patent document to be evaluated, and obtaining an adjustment result; according to the adjustment result, using each claim item of the patent document to be evaluated to traverse each similar patent document for technical feature comparison, and obtaining the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; A technical feature spectrum is drawn according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim item and each similar patent document, and the magnitude of the similarity is corresponding to a preset color tone. The higher the similarity, the greater the thermal value of the corresponding point on the spectrum, and the lower the similarity, the smaller the thermal value corresponding to the point on the spectrum, so as to achieve: from the overall color warmth and coldness, the overall overview of the technical novelty of the patent document to be evaluated can be intuitively seen; using the vertical color as a clue, the comparison documents or conflicting applications that may cause technical innovation conflicts can be found; using the horizontal color as a clue, the high-quality innovation points of the document to be applied can be found; According to the technical feature spectrum diagram, the patent document to be evaluated is pre-evaluated to obtain the pre-evaluation result of the patent document to be evaluated, including: when receiving the user's click operation on the intersection, obtaining the pre-evaluation result of comparing a single claim of the patent document to be evaluated with a single similar patent document; or, when receiving the user's click operation on the preset ordinate on the vertical axis, obtaining the pre-evaluation result of comparing a single claim of the patent document to be evaluated with all similar patent documents; or, when receiving the user's click operation on the preset abscissa on the horizontal axis, obtaining the pre-evaluation result of comparing all claims of the patent document to be evaluated with a single similar patent document; or, when receiving the user's click operation on the feature similarity extreme value in the technical feature spectrum diagram, obtaining the pre-evaluation result of comparing all claims of the patent document to be evaluated with the similar patent document with the greatest similarity corresponding to each claim.

2. The method for pre-evaluating a patent document according to claim 1, characterized in that: Using each claim item of the patent document to be evaluated to traverse each similar patent document to compare technical features, and obtain the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document, including: using each claim item of the patent document to be evaluated to traverse each similar patent document to compare the technical features of the entire document, and obtain the similarity of technical features with the greatest similarity between each claim item in the patent document to be evaluated and each similar patent document; A technical feature spectrum diagram is drawn according to the similarity of technical features between each claim item in the patent document to be evaluated and each similar patent document, including: drawing a technical feature spectrum diagram according to the similarity of technical features with the greatest similarity between each claim item in the patent document to be evaluated and each similar patent document; the intersection point of each claim item and each similar patent document in the technical feature spectrum diagram corresponds to the similarity of technical features with the greatest similarity between each claim item and each similar patent document.

3. The method for pre-evaluating a patent document according to claim 1, characterized in that: Input the information of the patent document to be evaluated into the pre-established semantic analysis model to retrieve multiple similar patent documents similar to the patent document to be evaluated, including: retrieving each similar patent document similar to the patent document to be evaluated according to the following method: Divide the patent document to be evaluated and each document to be compared into multiple parts; Performing semantic analysis on each part to obtain the semantic analysis result of each part; Determine the weight value of each part of the patent document to be evaluated and each part of each document to be compared according to the semantic analysis results of each part; According to the weight value of each part of the patent document to be evaluated and each document to be compared, a weighted average result of the patent document to be evaluated and each document to be compared is obtained; Determine the similarity between the patent document to be evaluated and the document to be compared based on the weighted average results of the patent document to be evaluated and each document to be compared; When the similarity between the patent document to be evaluated and the document to be compared is greater than a preset value, the document to be compared is regarded as an approximate patent document of the patent document to be evaluated.

4. The method for pre-evaluating a patent document according to claim 1, characterized in that: It also includes: conducting a preliminary assessment of the competitive situation and commercial prospects of the patent document to be assessed, and obtaining preliminary assessment results of the competitive situation and commercial prospects of the patent document to be assessed; Preliminary evaluation of the patent document to be evaluated is performed according to the technical feature spectrum to obtain a preliminary evaluation result of the patent document to be evaluated, including: preliminary evaluation of the novelty of the patent document to be evaluated is performed according to the technical feature spectrum to obtain a preliminary evaluation result of the novelty of the patent document to be evaluated; The patent document pre-evaluation method further includes: obtaining a comprehensive pre-evaluation result of the patent document to be evaluated based on the novelty pre-evaluation result of the patent document to be evaluated, as well as the pre-evaluation result of the competition situation and business prospects.

5. The method for pre-evaluating a patent document according to claim 1, characterized in that: Also includes: Adjust the patent application strategy of the patent document to be evaluated according to the preliminary evaluation results of the patent document to be evaluated.

6. A patent document pre-evaluation device, characterized in that: include: An intelligent retrieval unit, used to input the information of the patent document to be evaluated into a pre-established semantic analysis model, and retrieve multiple similar patent documents similar to the patent document to be evaluated; A traversal feature comparison unit is used to traverse each similar patent document using each claim item of the patent document to be evaluated to compare the technical features, and obtain the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document, which includes: adjusting the size of the comparison feature of each similar patent document according to the size of each claim item of the patent document to be evaluated, and obtaining an adjustment result; according to the adjustment result, traversing each similar patent document using each claim item of the patent document to be evaluated to compare the technical features, and obtaining the similarity of the technical features of each claim item of the patent document to be evaluated and each similar patent document; A spectrum drawing unit is used to draw a technical feature spectrum according to the similarity of technical features between each claim item of the patent document to be evaluated and each similar patent document; the vertical axis in the technical feature spectrum corresponds to multiple claims items of the patent document to be evaluated, and the horizontal axis corresponds to multiple similar patent documents; the intersection of each claim item and each similar patent document in the technical feature spectrum corresponds to the similarity of technical features between each claim item and each similar patent document, and the magnitude of the similarity is corresponding to a preset color tone. The higher the similarity, the greater the thermal value of the corresponding point on the spectrum, and the lower the similarity, the smaller the thermal value corresponding to the point on the spectrum, so as to achieve: from the overall color cold and warm conditions, the overall overview of the technical novelty of the patent document to be evaluated can be intuitively seen; using the vertical color as a clue, the comparison documents or conflicting applications that may cause technical innovation conflicts can be found; using the horizontal color as a clue, the high-quality innovation points of the document to be applied can be found; The visual interactive evaluation unit is used to pre-evaluate the patent document to be evaluated according to the technical feature spectrum diagram, and obtain the pre-evaluation result of the patent document to be evaluated, including: when receiving the user's click operation on the intersection, obtaining the pre-evaluation result of comparing a single claim of the patent document to be evaluated with a single similar patent document; or, when receiving the user's click operation on the preset vertical coordinate on the vertical axis, obtaining the pre-evaluation result of comparing a single claim of the patent document to be evaluated with all similar patent documents; or, when receiving the user's click operation on the preset horizontal coordinate on the horizontal axis, obtaining the pre-evaluation result of comparing all claims of the patent document to be evaluated with a single similar patent document; or, when receiving the user's click operation on the feature similarity extreme value in the technical feature spectrum diagram, obtaining the pre-evaluation result of comparing all claims of the patent document to be evaluated with the similar patent document with the maximum similarity corresponding to each claim.

7. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

9. A computer program product, characterized in that The computer program product comprises a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images